xref: /openbmc/qemu/tcg/s390x/tcg-target.c.inc (revision 07d5d502f2b4a8eedda3c6bdfcab31dc36d1d1d5)
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#include "elf.h"
28
29/* Used for function call generation. */
30#define TCG_TARGET_STACK_ALIGN          8
31#define TCG_TARGET_CALL_STACK_OFFSET    160
32#define TCG_TARGET_CALL_ARG_I32         TCG_CALL_ARG_EXTEND
33#define TCG_TARGET_CALL_ARG_I64         TCG_CALL_ARG_NORMAL
34#define TCG_TARGET_CALL_ARG_I128        TCG_CALL_ARG_BY_REF
35#define TCG_TARGET_CALL_RET_I128        TCG_CALL_RET_BY_REF
36
37#define TCG_CT_CONST_S16        (1 << 8)
38#define TCG_CT_CONST_S32        (1 << 9)
39#define TCG_CT_CONST_U32        (1 << 10)
40#define TCG_CT_CONST_ZERO       (1 << 11)
41#define TCG_CT_CONST_P32        (1 << 12)
42#define TCG_CT_CONST_INV        (1 << 13)
43#define TCG_CT_CONST_INVRISBG   (1 << 14)
44#define TCG_CT_CONST_CMP        (1 << 15)
45#define TCG_CT_CONST_M1         (1 << 16)
46
47#define ALL_GENERAL_REGS     MAKE_64BIT_MASK(0, 16)
48#define ALL_VECTOR_REGS      MAKE_64BIT_MASK(32, 32)
49
50/* Several places within the instruction set 0 means "no register"
51   rather than TCG_REG_R0.  */
52#define TCG_REG_NONE    0
53
54/* A scratch register that may be be used throughout the backend.  */
55#define TCG_TMP0        TCG_REG_R1
56#define TCG_VEC_TMP0    TCG_REG_V31
57
58#define TCG_GUEST_BASE_REG TCG_REG_R13
59
60/* All of the following instructions are prefixed with their instruction
61   format, and are defined as 8- or 16-bit quantities, even when the two
62   halves of the 16-bit quantity may appear 32 bits apart in the insn.
63   This makes it easy to copy the values from the tables in Appendix B.  */
64typedef enum S390Opcode {
65    RIL_AFI     = 0xc209,
66    RIL_AGFI    = 0xc208,
67    RIL_ALFI    = 0xc20b,
68    RIL_ALGFI   = 0xc20a,
69    RIL_BRASL   = 0xc005,
70    RIL_BRCL    = 0xc004,
71    RIL_CFI     = 0xc20d,
72    RIL_CGFI    = 0xc20c,
73    RIL_CLFI    = 0xc20f,
74    RIL_CLGFI   = 0xc20e,
75    RIL_CLRL    = 0xc60f,
76    RIL_CLGRL   = 0xc60a,
77    RIL_CRL     = 0xc60d,
78    RIL_CGRL    = 0xc608,
79    RIL_IIHF    = 0xc008,
80    RIL_IILF    = 0xc009,
81    RIL_LARL    = 0xc000,
82    RIL_LGFI    = 0xc001,
83    RIL_LGRL    = 0xc408,
84    RIL_LLIHF   = 0xc00e,
85    RIL_LLILF   = 0xc00f,
86    RIL_LRL     = 0xc40d,
87    RIL_MSFI    = 0xc201,
88    RIL_MSGFI   = 0xc200,
89    RIL_NIHF    = 0xc00a,
90    RIL_NILF    = 0xc00b,
91    RIL_OIHF    = 0xc00c,
92    RIL_OILF    = 0xc00d,
93    RIL_SLFI    = 0xc205,
94    RIL_SLGFI   = 0xc204,
95    RIL_XIHF    = 0xc006,
96    RIL_XILF    = 0xc007,
97
98    RI_AGHI     = 0xa70b,
99    RI_AHI      = 0xa70a,
100    RI_BRC      = 0xa704,
101    RI_CHI      = 0xa70e,
102    RI_CGHI     = 0xa70f,
103    RI_IIHH     = 0xa500,
104    RI_IIHL     = 0xa501,
105    RI_IILH     = 0xa502,
106    RI_IILL     = 0xa503,
107    RI_LGHI     = 0xa709,
108    RI_LLIHH    = 0xa50c,
109    RI_LLIHL    = 0xa50d,
110    RI_LLILH    = 0xa50e,
111    RI_LLILL    = 0xa50f,
112    RI_MGHI     = 0xa70d,
113    RI_MHI      = 0xa70c,
114    RI_NIHH     = 0xa504,
115    RI_NIHL     = 0xa505,
116    RI_NILH     = 0xa506,
117    RI_NILL     = 0xa507,
118    RI_OIHH     = 0xa508,
119    RI_OIHL     = 0xa509,
120    RI_OILH     = 0xa50a,
121    RI_OILL     = 0xa50b,
122    RI_TMLL     = 0xa701,
123    RI_TMLH     = 0xa700,
124    RI_TMHL     = 0xa703,
125    RI_TMHH     = 0xa702,
126
127    RIEb_CGRJ    = 0xec64,
128    RIEb_CLGRJ   = 0xec65,
129    RIEb_CLRJ    = 0xec77,
130    RIEb_CRJ     = 0xec76,
131
132    RIEc_CGIJ    = 0xec7c,
133    RIEc_CIJ     = 0xec7e,
134    RIEc_CLGIJ   = 0xec7d,
135    RIEc_CLIJ    = 0xec7f,
136
137    RIEf_RISBG   = 0xec55,
138
139    RIEg_LOCGHI  = 0xec46,
140
141    RRE_AGR     = 0xb908,
142    RRE_ALGR    = 0xb90a,
143    RRE_ALCR    = 0xb998,
144    RRE_ALCGR   = 0xb988,
145    RRE_ALGFR   = 0xb91a,
146    RRE_CGR     = 0xb920,
147    RRE_CLGR    = 0xb921,
148    RRE_DLGR    = 0xb987,
149    RRE_DLR     = 0xb997,
150    RRE_DSGFR   = 0xb91d,
151    RRE_DSGR    = 0xb90d,
152    RRE_FLOGR   = 0xb983,
153    RRE_LGBR    = 0xb906,
154    RRE_LCGR    = 0xb903,
155    RRE_LGFR    = 0xb914,
156    RRE_LGHR    = 0xb907,
157    RRE_LGR     = 0xb904,
158    RRE_LLGCR   = 0xb984,
159    RRE_LLGFR   = 0xb916,
160    RRE_LLGHR   = 0xb985,
161    RRE_LRVR    = 0xb91f,
162    RRE_LRVGR   = 0xb90f,
163    RRE_LTGR    = 0xb902,
164    RRE_MLGR    = 0xb986,
165    RRE_MSGR    = 0xb90c,
166    RRE_MSR     = 0xb252,
167    RRE_NGR     = 0xb980,
168    RRE_OGR     = 0xb981,
169    RRE_SGR     = 0xb909,
170    RRE_SLGR    = 0xb90b,
171    RRE_SLBR    = 0xb999,
172    RRE_SLBGR   = 0xb989,
173    RRE_XGR     = 0xb982,
174
175    RRFa_MGRK   = 0xb9ec,
176    RRFa_MSRKC  = 0xb9fd,
177    RRFa_MSGRKC = 0xb9ed,
178    RRFa_NCRK   = 0xb9f5,
179    RRFa_NCGRK  = 0xb9e5,
180    RRFa_NNRK   = 0xb974,
181    RRFa_NNGRK  = 0xb964,
182    RRFa_NORK   = 0xb976,
183    RRFa_NOGRK  = 0xb966,
184    RRFa_NRK    = 0xb9f4,
185    RRFa_NGRK   = 0xb9e4,
186    RRFa_NXRK   = 0xb977,
187    RRFa_NXGRK  = 0xb967,
188    RRFa_OCRK   = 0xb975,
189    RRFa_OCGRK  = 0xb965,
190    RRFa_ORK    = 0xb9f6,
191    RRFa_OGRK   = 0xb9e6,
192    RRFa_SRK    = 0xb9f9,
193    RRFa_SGRK   = 0xb9e9,
194    RRFa_SLRK   = 0xb9fb,
195    RRFa_SLGRK  = 0xb9eb,
196    RRFa_XRK    = 0xb9f7,
197    RRFa_XGRK   = 0xb9e7,
198
199    RRFam_SELGR = 0xb9e3,
200
201    RRFc_LOCR   = 0xb9f2,
202    RRFc_LOCGR  = 0xb9e2,
203    RRFc_POPCNT = 0xb9e1,
204
205    RR_AR       = 0x1a,
206    RR_ALR      = 0x1e,
207    RR_BASR     = 0x0d,
208    RR_BCR      = 0x07,
209    RR_CLR      = 0x15,
210    RR_CR       = 0x19,
211    RR_DR       = 0x1d,
212    RR_LCR      = 0x13,
213    RR_LR       = 0x18,
214    RR_LTR      = 0x12,
215    RR_NR       = 0x14,
216    RR_OR       = 0x16,
217    RR_SR       = 0x1b,
218    RR_SLR      = 0x1f,
219    RR_XR       = 0x17,
220
221    RSY_RLL     = 0xeb1d,
222    RSY_RLLG    = 0xeb1c,
223    RSY_SLLG    = 0xeb0d,
224    RSY_SLLK    = 0xebdf,
225    RSY_SRAG    = 0xeb0a,
226    RSY_SRAK    = 0xebdc,
227    RSY_SRLG    = 0xeb0c,
228    RSY_SRLK    = 0xebde,
229
230    RS_SLL      = 0x89,
231    RS_SRA      = 0x8a,
232    RS_SRL      = 0x88,
233
234    RXY_AG      = 0xe308,
235    RXY_AY      = 0xe35a,
236    RXY_CG      = 0xe320,
237    RXY_CLG     = 0xe321,
238    RXY_CLY     = 0xe355,
239    RXY_CY      = 0xe359,
240    RXY_LAY     = 0xe371,
241    RXY_LB      = 0xe376,
242    RXY_LG      = 0xe304,
243    RXY_LGB     = 0xe377,
244    RXY_LGF     = 0xe314,
245    RXY_LGH     = 0xe315,
246    RXY_LHY     = 0xe378,
247    RXY_LLGC    = 0xe390,
248    RXY_LLGF    = 0xe316,
249    RXY_LLGH    = 0xe391,
250    RXY_LMG     = 0xeb04,
251    RXY_LPQ     = 0xe38f,
252    RXY_LRV     = 0xe31e,
253    RXY_LRVG    = 0xe30f,
254    RXY_LRVH    = 0xe31f,
255    RXY_LY      = 0xe358,
256    RXY_NG      = 0xe380,
257    RXY_OG      = 0xe381,
258    RXY_STCY    = 0xe372,
259    RXY_STG     = 0xe324,
260    RXY_STHY    = 0xe370,
261    RXY_STMG    = 0xeb24,
262    RXY_STPQ    = 0xe38e,
263    RXY_STRV    = 0xe33e,
264    RXY_STRVG   = 0xe32f,
265    RXY_STRVH   = 0xe33f,
266    RXY_STY     = 0xe350,
267    RXY_XG      = 0xe382,
268
269    RX_A        = 0x5a,
270    RX_C        = 0x59,
271    RX_L        = 0x58,
272    RX_LA       = 0x41,
273    RX_LH       = 0x48,
274    RX_ST       = 0x50,
275    RX_STC      = 0x42,
276    RX_STH      = 0x40,
277
278    VRIa_VGBM   = 0xe744,
279    VRIa_VREPI  = 0xe745,
280    VRIb_VGM    = 0xe746,
281    VRIc_VREP   = 0xe74d,
282
283    VRRa_VLC    = 0xe7de,
284    VRRa_VLP    = 0xe7df,
285    VRRa_VLR    = 0xe756,
286    VRRc_VA     = 0xe7f3,
287    VRRc_VCEQ   = 0xe7f8,   /* we leave the m5 cs field 0 */
288    VRRc_VCH    = 0xe7fb,   /* " */
289    VRRc_VCHL   = 0xe7f9,   /* " */
290    VRRc_VERLLV = 0xe773,
291    VRRc_VESLV  = 0xe770,
292    VRRc_VESRAV = 0xe77a,
293    VRRc_VESRLV = 0xe778,
294    VRRc_VML    = 0xe7a2,
295    VRRc_VMN    = 0xe7fe,
296    VRRc_VMNL   = 0xe7fc,
297    VRRc_VMX    = 0xe7ff,
298    VRRc_VMXL   = 0xe7fd,
299    VRRc_VN     = 0xe768,
300    VRRc_VNC    = 0xe769,
301    VRRc_VNN    = 0xe76e,
302    VRRc_VNO    = 0xe76b,
303    VRRc_VNX    = 0xe76c,
304    VRRc_VO     = 0xe76a,
305    VRRc_VOC    = 0xe76f,
306    VRRc_VPKS   = 0xe797,   /* we leave the m5 cs field 0 */
307    VRRc_VS     = 0xe7f7,
308    VRRa_VUPH   = 0xe7d7,
309    VRRa_VUPL   = 0xe7d6,
310    VRRc_VX     = 0xe76d,
311    VRRe_VSEL   = 0xe78d,
312    VRRf_VLVGP  = 0xe762,
313
314    VRSa_VERLL  = 0xe733,
315    VRSa_VESL   = 0xe730,
316    VRSa_VESRA  = 0xe73a,
317    VRSa_VESRL  = 0xe738,
318    VRSb_VLVG   = 0xe722,
319    VRSc_VLGV   = 0xe721,
320
321    VRX_VL      = 0xe706,
322    VRX_VLLEZ   = 0xe704,
323    VRX_VLREP   = 0xe705,
324    VRX_VST     = 0xe70e,
325    VRX_VSTEF   = 0xe70b,
326    VRX_VSTEG   = 0xe70a,
327
328    NOP         = 0x0707,
329} S390Opcode;
330
331#ifdef CONFIG_DEBUG_TCG
332static const char * const tcg_target_reg_names[TCG_TARGET_NB_REGS] = {
333    "%r0",  "%r1",  "%r2",  "%r3",  "%r4",  "%r5",  "%r6",  "%r7",
334    "%r8",  "%r9",  "%r10", "%r11", "%r12", "%r13", "%r14", "%r15",
335    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
336    "%v0",  "%v1",  "%v2",  "%v3",  "%v4",  "%v5",  "%v6",  "%v7",
337    "%v8",  "%v9",  "%v10", "%v11", "%v12", "%v13", "%v14", "%v15",
338    "%v16", "%v17", "%v18", "%v19", "%v20", "%v21", "%v22", "%v23",
339    "%v24", "%v25", "%v26", "%v27", "%v28", "%v29", "%v30", "%v31",
340};
341#endif
342
343/* Since R6 is a potential argument register, choose it last of the
344   call-saved registers.  Likewise prefer the call-clobbered registers
345   in reverse order to maximize the chance of avoiding the arguments.  */
346static const int tcg_target_reg_alloc_order[] = {
347    /* Call saved registers.  */
348    TCG_REG_R13,
349    TCG_REG_R12,
350    TCG_REG_R11,
351    TCG_REG_R10,
352    TCG_REG_R9,
353    TCG_REG_R8,
354    TCG_REG_R7,
355    TCG_REG_R6,
356    /* Call clobbered registers.  */
357    TCG_REG_R14,
358    TCG_REG_R0,
359    TCG_REG_R1,
360    /* Argument registers, in reverse order of allocation.  */
361    TCG_REG_R5,
362    TCG_REG_R4,
363    TCG_REG_R3,
364    TCG_REG_R2,
365
366    /* V8-V15 are call saved, and omitted. */
367    TCG_REG_V0,
368    TCG_REG_V1,
369    TCG_REG_V2,
370    TCG_REG_V3,
371    TCG_REG_V4,
372    TCG_REG_V5,
373    TCG_REG_V6,
374    TCG_REG_V7,
375    TCG_REG_V16,
376    TCG_REG_V17,
377    TCG_REG_V18,
378    TCG_REG_V19,
379    TCG_REG_V20,
380    TCG_REG_V21,
381    TCG_REG_V22,
382    TCG_REG_V23,
383    TCG_REG_V24,
384    TCG_REG_V25,
385    TCG_REG_V26,
386    TCG_REG_V27,
387    TCG_REG_V28,
388    TCG_REG_V29,
389    TCG_REG_V30,
390    TCG_REG_V31,
391};
392
393static const int tcg_target_call_iarg_regs[] = {
394    TCG_REG_R2,
395    TCG_REG_R3,
396    TCG_REG_R4,
397    TCG_REG_R5,
398    TCG_REG_R6,
399};
400
401static TCGReg tcg_target_call_oarg_reg(TCGCallReturnKind kind, int slot)
402{
403    tcg_debug_assert(kind == TCG_CALL_RET_NORMAL);
404    tcg_debug_assert(slot == 0);
405    return TCG_REG_R2;
406}
407
408#define S390_CC_EQ      8
409#define S390_CC_LT      4
410#define S390_CC_GT      2
411#define S390_CC_OV      1
412#define S390_CC_NE      (S390_CC_LT | S390_CC_GT)
413#define S390_CC_LE      (S390_CC_LT | S390_CC_EQ)
414#define S390_CC_GE      (S390_CC_GT | S390_CC_EQ)
415#define S390_CC_NEVER   0
416#define S390_CC_ALWAYS  15
417
418#define S390_TM_EQ      8  /* CC == 0 */
419#define S390_TM_NE      7  /* CC in {1,2,3} */
420
421/* Condition codes that result from a COMPARE and COMPARE LOGICAL.  */
422static const uint8_t tcg_cond_to_s390_cond[16] = {
423    [TCG_COND_EQ]  = S390_CC_EQ,
424    [TCG_COND_NE]  = S390_CC_NE,
425    [TCG_COND_TSTEQ] = S390_CC_EQ,
426    [TCG_COND_TSTNE] = S390_CC_NE,
427    [TCG_COND_LT]  = S390_CC_LT,
428    [TCG_COND_LE]  = S390_CC_LE,
429    [TCG_COND_GT]  = S390_CC_GT,
430    [TCG_COND_GE]  = S390_CC_GE,
431    [TCG_COND_LTU] = S390_CC_LT,
432    [TCG_COND_LEU] = S390_CC_LE,
433    [TCG_COND_GTU] = S390_CC_GT,
434    [TCG_COND_GEU] = S390_CC_GE,
435};
436
437/* Condition codes that result from a LOAD AND TEST.  Here, we have no
438   unsigned instruction variation, however since the test is vs zero we
439   can re-map the outcomes appropriately.  */
440static const uint8_t tcg_cond_to_ltr_cond[16] = {
441    [TCG_COND_EQ]  = S390_CC_EQ,
442    [TCG_COND_NE]  = S390_CC_NE,
443    [TCG_COND_TSTEQ] = S390_CC_ALWAYS,
444    [TCG_COND_TSTNE] = S390_CC_NEVER,
445    [TCG_COND_LT]  = S390_CC_LT,
446    [TCG_COND_LE]  = S390_CC_LE,
447    [TCG_COND_GT]  = S390_CC_GT,
448    [TCG_COND_GE]  = S390_CC_GE,
449    [TCG_COND_LTU] = S390_CC_NEVER,
450    [TCG_COND_LEU] = S390_CC_EQ,
451    [TCG_COND_GTU] = S390_CC_NE,
452    [TCG_COND_GEU] = S390_CC_ALWAYS,
453};
454
455static const tcg_insn_unit *tb_ret_addr;
456uint64_t s390_facilities[3];
457
458static inline bool is_general_reg(TCGReg r)
459{
460    return r <= TCG_REG_R15;
461}
462
463static inline bool is_vector_reg(TCGReg r)
464{
465    return r >= TCG_REG_V0 && r <= TCG_REG_V31;
466}
467
468static bool patch_reloc(tcg_insn_unit *src_rw, int type,
469                        intptr_t value, intptr_t addend)
470{
471    const tcg_insn_unit *src_rx = tcg_splitwx_to_rx(src_rw);
472    intptr_t pcrel2;
473    uint32_t old;
474
475    value += addend;
476    pcrel2 = (tcg_insn_unit *)value - src_rx;
477
478    switch (type) {
479    case R_390_PC16DBL:
480        if (pcrel2 == (int16_t)pcrel2) {
481            tcg_patch16(src_rw, pcrel2);
482            return true;
483        }
484        break;
485    case R_390_PC32DBL:
486        if (pcrel2 == (int32_t)pcrel2) {
487            tcg_patch32(src_rw, pcrel2);
488            return true;
489        }
490        break;
491    case R_390_20:
492        if (value == sextract64(value, 0, 20)) {
493            old = *(uint32_t *)src_rw & 0xf00000ff;
494            old |= ((value & 0xfff) << 16) | ((value & 0xff000) >> 4);
495            tcg_patch32(src_rw, old);
496            return true;
497        }
498        break;
499    default:
500        g_assert_not_reached();
501    }
502    return false;
503}
504
505static int is_const_p16(uint64_t val)
506{
507    for (int i = 0; i < 4; ++i) {
508        uint64_t mask = 0xffffull << (i * 16);
509        if ((val & ~mask) == 0) {
510            return i;
511        }
512    }
513    return -1;
514}
515
516static int is_const_p32(uint64_t val)
517{
518    if ((val & 0xffffffff00000000ull) == 0) {
519        return 0;
520    }
521    if ((val & 0x00000000ffffffffull) == 0) {
522        return 1;
523    }
524    return -1;
525}
526
527/*
528 * Accept bit patterns like these:
529 *  0....01....1
530 *  1....10....0
531 *  1..10..01..1
532 *  0..01..10..0
533 * Copied from gcc sources.
534 */
535static bool risbg_mask(uint64_t c)
536{
537    uint64_t lsb;
538    /* We don't change the number of transitions by inverting,
539       so make sure we start with the LSB zero.  */
540    if (c & 1) {
541        c = ~c;
542    }
543    /* Reject all zeros or all ones.  */
544    if (c == 0) {
545        return false;
546    }
547    /* Find the first transition.  */
548    lsb = c & -c;
549    /* Invert to look for a second transition.  */
550    c = ~c;
551    /* Erase the first transition.  */
552    c &= -lsb;
553    /* Find the second transition, if any.  */
554    lsb = c & -c;
555    /* Match if all the bits are 1's, or if c is zero.  */
556    return c == -lsb;
557}
558
559/* Test if a constant matches the constraint. */
560static bool tcg_target_const_match(int64_t val, int ct,
561                                   TCGType type, TCGCond cond, int vece)
562{
563    uint64_t uval = val;
564
565    if (ct & TCG_CT_CONST) {
566        return true;
567    }
568    if (type == TCG_TYPE_I32) {
569        uval = (uint32_t)val;
570        val = (int32_t)val;
571    }
572
573    if (ct & TCG_CT_CONST_CMP) {
574        if (is_tst_cond(cond)) {
575            if (is_const_p16(uval) >= 0) {
576                return true;  /* TMxx */
577            }
578            if (risbg_mask(uval)) {
579                return true;  /* RISBG */
580            }
581            return false;
582        }
583
584        if (type == TCG_TYPE_I32) {
585            return true;
586        }
587
588        switch (cond) {
589        case TCG_COND_EQ:
590        case TCG_COND_NE:
591            ct |= TCG_CT_CONST_S32 | TCG_CT_CONST_U32;  /* CGFI or CLGFI */
592            break;
593        case TCG_COND_LT:
594        case TCG_COND_GE:
595        case TCG_COND_LE:
596        case TCG_COND_GT:
597            ct |= TCG_CT_CONST_S32;  /* CGFI */
598            break;
599        case TCG_COND_LTU:
600        case TCG_COND_GEU:
601        case TCG_COND_LEU:
602        case TCG_COND_GTU:
603            ct |= TCG_CT_CONST_U32;  /* CLGFI */
604            break;
605        case TCG_COND_TSTNE:
606        case TCG_COND_TSTEQ:
607            /* checked above, fallthru */
608        default:
609            g_assert_not_reached();
610        }
611    }
612
613    if ((ct & TCG_CT_CONST_S32) && val == (int32_t)val) {
614        return true;
615    }
616    if ((ct & TCG_CT_CONST_U32) && val == (uint32_t)val) {
617        return true;
618    }
619    if ((ct & TCG_CT_CONST_S16) && val == (int16_t)val) {
620        return true;
621    }
622    if ((ct & TCG_CT_CONST_ZERO) && val == 0) {
623        return true;
624    }
625    if ((ct & TCG_CT_CONST_M1) && val == -1) {
626        return true;
627    }
628
629    if (ct & TCG_CT_CONST_INV) {
630        val = ~val;
631    }
632    if ((ct & TCG_CT_CONST_P32) && is_const_p32(val) >= 0) {
633        return true;
634    }
635    if ((ct & TCG_CT_CONST_INVRISBG) && risbg_mask(~val)) {
636        return true;
637    }
638    return false;
639}
640
641/* Emit instructions according to the given instruction format.  */
642
643static void tcg_out_insn_RR(TCGContext *s, S390Opcode op, TCGReg r1, TCGReg r2)
644{
645    tcg_out16(s, (op << 8) | (r1 << 4) | r2);
646}
647
648static void tcg_out_insn_RRE(TCGContext *s, S390Opcode op,
649                             TCGReg r1, TCGReg r2)
650{
651    tcg_out32(s, (op << 16) | (r1 << 4) | r2);
652}
653
654/* RRF-a without the m4 field */
655static void tcg_out_insn_RRFa(TCGContext *s, S390Opcode op,
656                              TCGReg r1, TCGReg r2, TCGReg r3)
657{
658    tcg_out32(s, (op << 16) | (r3 << 12) | (r1 << 4) | r2);
659}
660
661/* RRF-a with the m4 field */
662static void tcg_out_insn_RRFam(TCGContext *s, S390Opcode op,
663                               TCGReg r1, TCGReg r2, TCGReg r3, int m4)
664{
665    tcg_out32(s, (op << 16) | (r3 << 12) | (m4 << 8) | (r1 << 4) | r2);
666}
667
668static void tcg_out_insn_RRFc(TCGContext *s, S390Opcode op,
669                              TCGReg r1, TCGReg r2, int m3)
670{
671    tcg_out32(s, (op << 16) | (m3 << 12) | (r1 << 4) | r2);
672}
673
674static void tcg_out_insn_RI(TCGContext *s, S390Opcode op, TCGReg r1, int i2)
675{
676    tcg_out32(s, (op << 16) | (r1 << 20) | (i2 & 0xffff));
677}
678
679static void tcg_out_insn_RIEg(TCGContext *s, S390Opcode op, TCGReg r1,
680                             int i2, int m3)
681{
682    tcg_out16(s, (op & 0xff00) | (r1 << 4) | m3);
683    tcg_out32(s, (i2 << 16) | (op & 0xff));
684}
685
686static void tcg_out_insn_RIL(TCGContext *s, S390Opcode op, TCGReg r1, int i2)
687{
688    tcg_out16(s, op | (r1 << 4));
689    tcg_out32(s, i2);
690}
691
692static void tcg_out_insn_RS(TCGContext *s, S390Opcode op, TCGReg r1,
693                            TCGReg b2, TCGReg r3, int disp)
694{
695    tcg_out32(s, (op << 24) | (r1 << 20) | (r3 << 16) | (b2 << 12)
696              | (disp & 0xfff));
697}
698
699static void tcg_out_insn_RSY(TCGContext *s, S390Opcode op, TCGReg r1,
700                             TCGReg b2, TCGReg r3, int disp)
701{
702    tcg_out16(s, (op & 0xff00) | (r1 << 4) | r3);
703    tcg_out32(s, (op & 0xff) | (b2 << 28)
704              | ((disp & 0xfff) << 16) | ((disp & 0xff000) >> 4));
705}
706
707#define tcg_out_insn_RX   tcg_out_insn_RS
708#define tcg_out_insn_RXY  tcg_out_insn_RSY
709
710static int RXB(TCGReg v1, TCGReg v2, TCGReg v3, TCGReg v4)
711{
712    /*
713     * Shift bit 4 of each regno to its corresponding bit of RXB.
714     * RXB itself begins at bit 8 of the instruction so 8 - 4 = 4
715     * is the left-shift of the 4th operand.
716     */
717    return ((v1 & 0x10) << (4 + 3))
718         | ((v2 & 0x10) << (4 + 2))
719         | ((v3 & 0x10) << (4 + 1))
720         | ((v4 & 0x10) << (4 + 0));
721}
722
723static void tcg_out_insn_VRIa(TCGContext *s, S390Opcode op,
724                              TCGReg v1, uint16_t i2, int m3)
725{
726    tcg_debug_assert(is_vector_reg(v1));
727    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4));
728    tcg_out16(s, i2);
729    tcg_out16(s, (op & 0x00ff) | RXB(v1, 0, 0, 0) | (m3 << 12));
730}
731
732static void tcg_out_insn_VRIb(TCGContext *s, S390Opcode op,
733                              TCGReg v1, uint8_t i2, uint8_t i3, int m4)
734{
735    tcg_debug_assert(is_vector_reg(v1));
736    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4));
737    tcg_out16(s, (i2 << 8) | (i3 & 0xff));
738    tcg_out16(s, (op & 0x00ff) | RXB(v1, 0, 0, 0) | (m4 << 12));
739}
740
741static void tcg_out_insn_VRIc(TCGContext *s, S390Opcode op,
742                              TCGReg v1, uint16_t i2, TCGReg v3, int m4)
743{
744    tcg_debug_assert(is_vector_reg(v1));
745    tcg_debug_assert(is_vector_reg(v3));
746    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4) | (v3 & 0xf));
747    tcg_out16(s, i2);
748    tcg_out16(s, (op & 0x00ff) | RXB(v1, v3, 0, 0) | (m4 << 12));
749}
750
751static void tcg_out_insn_VRRa(TCGContext *s, S390Opcode op,
752                              TCGReg v1, TCGReg v2, int m3)
753{
754    tcg_debug_assert(is_vector_reg(v1));
755    tcg_debug_assert(is_vector_reg(v2));
756    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4) | (v2 & 0xf));
757    tcg_out32(s, (op & 0x00ff) | RXB(v1, v2, 0, 0) | (m3 << 12));
758}
759
760static void tcg_out_insn_VRRc(TCGContext *s, S390Opcode op,
761                              TCGReg v1, TCGReg v2, TCGReg v3, int m4)
762{
763    tcg_debug_assert(is_vector_reg(v1));
764    tcg_debug_assert(is_vector_reg(v2));
765    tcg_debug_assert(is_vector_reg(v3));
766    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4) | (v2 & 0xf));
767    tcg_out16(s, v3 << 12);
768    tcg_out16(s, (op & 0x00ff) | RXB(v1, v2, v3, 0) | (m4 << 12));
769}
770
771static void tcg_out_insn_VRRe(TCGContext *s, S390Opcode op,
772                              TCGReg v1, TCGReg v2, TCGReg v3, TCGReg v4)
773{
774    tcg_debug_assert(is_vector_reg(v1));
775    tcg_debug_assert(is_vector_reg(v2));
776    tcg_debug_assert(is_vector_reg(v3));
777    tcg_debug_assert(is_vector_reg(v4));
778    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4) | (v2 & 0xf));
779    tcg_out16(s, v3 << 12);
780    tcg_out16(s, (op & 0x00ff) | RXB(v1, v2, v3, v4) | (v4 << 12));
781}
782
783static void tcg_out_insn_VRRf(TCGContext *s, S390Opcode op,
784                              TCGReg v1, TCGReg r2, TCGReg r3)
785{
786    tcg_debug_assert(is_vector_reg(v1));
787    tcg_debug_assert(is_general_reg(r2));
788    tcg_debug_assert(is_general_reg(r3));
789    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4) | r2);
790    tcg_out16(s, r3 << 12);
791    tcg_out16(s, (op & 0x00ff) | RXB(v1, 0, 0, 0));
792}
793
794static void tcg_out_insn_VRSa(TCGContext *s, S390Opcode op, TCGReg v1,
795                              intptr_t d2, TCGReg b2, TCGReg v3, int m4)
796{
797    tcg_debug_assert(is_vector_reg(v1));
798    tcg_debug_assert(d2 >= 0 && d2 <= 0xfff);
799    tcg_debug_assert(is_general_reg(b2));
800    tcg_debug_assert(is_vector_reg(v3));
801    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4) | (v3 & 0xf));
802    tcg_out16(s, b2 << 12 | d2);
803    tcg_out16(s, (op & 0x00ff) | RXB(v1, v3, 0, 0) | (m4 << 12));
804}
805
806static void tcg_out_insn_VRSb(TCGContext *s, S390Opcode op, TCGReg v1,
807                              intptr_t d2, TCGReg b2, TCGReg r3, int m4)
808{
809    tcg_debug_assert(is_vector_reg(v1));
810    tcg_debug_assert(d2 >= 0 && d2 <= 0xfff);
811    tcg_debug_assert(is_general_reg(b2));
812    tcg_debug_assert(is_general_reg(r3));
813    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4) | r3);
814    tcg_out16(s, b2 << 12 | d2);
815    tcg_out16(s, (op & 0x00ff) | RXB(v1, 0, 0, 0) | (m4 << 12));
816}
817
818static void tcg_out_insn_VRSc(TCGContext *s, S390Opcode op, TCGReg r1,
819                              intptr_t d2, TCGReg b2, TCGReg v3, int m4)
820{
821    tcg_debug_assert(is_general_reg(r1));
822    tcg_debug_assert(d2 >= 0 && d2 <= 0xfff);
823    tcg_debug_assert(is_general_reg(b2));
824    tcg_debug_assert(is_vector_reg(v3));
825    tcg_out16(s, (op & 0xff00) | (r1 << 4) | (v3 & 0xf));
826    tcg_out16(s, b2 << 12 | d2);
827    tcg_out16(s, (op & 0x00ff) | RXB(0, v3, 0, 0) | (m4 << 12));
828}
829
830static void tcg_out_insn_VRX(TCGContext *s, S390Opcode op, TCGReg v1,
831                             TCGReg b2, TCGReg x2, intptr_t d2, int m3)
832{
833    tcg_debug_assert(is_vector_reg(v1));
834    tcg_debug_assert(d2 >= 0 && d2 <= 0xfff);
835    tcg_debug_assert(is_general_reg(x2));
836    tcg_debug_assert(is_general_reg(b2));
837    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4) | x2);
838    tcg_out16(s, (b2 << 12) | d2);
839    tcg_out16(s, (op & 0x00ff) | RXB(v1, 0, 0, 0) | (m3 << 12));
840}
841
842/* Emit an opcode with "type-checking" of the format.  */
843#define tcg_out_insn(S, FMT, OP, ...) \
844    glue(tcg_out_insn_,FMT)(S, glue(glue(FMT,_),OP), ## __VA_ARGS__)
845
846
847/* emit 64-bit shifts */
848static void tcg_out_sh64(TCGContext* s, S390Opcode op, TCGReg dest,
849                         TCGReg src, TCGReg sh_reg, int sh_imm)
850{
851    tcg_out_insn_RSY(s, op, dest, sh_reg, src, sh_imm);
852}
853
854/* emit 32-bit shifts */
855static void tcg_out_sh32(TCGContext* s, S390Opcode op, TCGReg dest,
856                         TCGReg sh_reg, int sh_imm)
857{
858    tcg_out_insn_RS(s, op, dest, sh_reg, 0, sh_imm);
859}
860
861static bool tcg_out_mov(TCGContext *s, TCGType type, TCGReg dst, TCGReg src)
862{
863    if (src == dst) {
864        return true;
865    }
866    switch (type) {
867    case TCG_TYPE_I32:
868        if (likely(is_general_reg(dst) && is_general_reg(src))) {
869            tcg_out_insn(s, RR, LR, dst, src);
870            break;
871        }
872        /* fallthru */
873
874    case TCG_TYPE_I64:
875        if (likely(is_general_reg(dst))) {
876            if (likely(is_general_reg(src))) {
877                tcg_out_insn(s, RRE, LGR, dst, src);
878            } else {
879                tcg_out_insn(s, VRSc, VLGV, dst, 0, 0, src, 3);
880            }
881            break;
882        } else if (is_general_reg(src)) {
883            tcg_out_insn(s, VRSb, VLVG, dst, 0, 0, src, 3);
884            break;
885        }
886        /* fallthru */
887
888    case TCG_TYPE_V64:
889    case TCG_TYPE_V128:
890        tcg_out_insn(s, VRRa, VLR, dst, src, 0);
891        break;
892
893    default:
894        g_assert_not_reached();
895    }
896    return true;
897}
898
899static const S390Opcode li_insns[4] = {
900    RI_LLILL, RI_LLILH, RI_LLIHL, RI_LLIHH
901};
902static const S390Opcode oi_insns[4] = {
903    RI_OILL, RI_OILH, RI_OIHL, RI_OIHH
904};
905static const S390Opcode lif_insns[2] = {
906    RIL_LLILF, RIL_LLIHF,
907};
908static const S390Opcode tm_insns[4] = {
909    RI_TMLL, RI_TMLH, RI_TMHL, RI_TMHH
910};
911
912/* load a register with an immediate value */
913static void tcg_out_movi(TCGContext *s, TCGType type,
914                         TCGReg ret, tcg_target_long sval)
915{
916    tcg_target_ulong uval = sval;
917    ptrdiff_t pc_off;
918    int i;
919
920    if (type == TCG_TYPE_I32) {
921        uval = (uint32_t)sval;
922        sval = (int32_t)sval;
923    }
924
925    /* Try all 32-bit insns that can load it in one go.  */
926    if (sval >= -0x8000 && sval < 0x8000) {
927        tcg_out_insn(s, RI, LGHI, ret, sval);
928        return;
929    }
930
931    i = is_const_p16(uval);
932    if (i >= 0) {
933        tcg_out_insn_RI(s, li_insns[i], ret, uval >> (i * 16));
934        return;
935    }
936
937    /* Try all 48-bit insns that can load it in one go.  */
938    if (sval == (int32_t)sval) {
939        tcg_out_insn(s, RIL, LGFI, ret, sval);
940        return;
941    }
942
943    i = is_const_p32(uval);
944    if (i >= 0) {
945        tcg_out_insn_RIL(s, lif_insns[i], ret, uval >> (i * 32));
946        return;
947    }
948
949    /* Try for PC-relative address load.  For odd addresses, add one. */
950    pc_off = tcg_pcrel_diff(s, (void *)sval) >> 1;
951    if (pc_off == (int32_t)pc_off) {
952        tcg_out_insn(s, RIL, LARL, ret, pc_off);
953        if (sval & 1) {
954            tcg_out_insn(s, RI, AGHI, ret, 1);
955        }
956        return;
957    }
958
959    /* Otherwise, load it by parts. */
960    i = is_const_p16((uint32_t)uval);
961    if (i >= 0) {
962        tcg_out_insn_RI(s, li_insns[i], ret, uval >> (i * 16));
963    } else {
964        tcg_out_insn(s, RIL, LLILF, ret, uval);
965    }
966    uval >>= 32;
967    i = is_const_p16(uval);
968    if (i >= 0) {
969        tcg_out_insn_RI(s, oi_insns[i + 2], ret, uval >> (i * 16));
970    } else {
971        tcg_out_insn(s, RIL, OIHF, ret, uval);
972    }
973}
974
975/* Emit a load/store type instruction.  Inputs are:
976   DATA:     The register to be loaded or stored.
977   BASE+OFS: The effective address.
978   OPC_RX:   If the operation has an RX format opcode (e.g. STC), otherwise 0.
979   OPC_RXY:  The RXY format opcode for the operation (e.g. STCY).  */
980
981static void tcg_out_mem(TCGContext *s, S390Opcode opc_rx, S390Opcode opc_rxy,
982                        TCGReg data, TCGReg base, TCGReg index,
983                        tcg_target_long ofs)
984{
985    if (ofs < -0x80000 || ofs >= 0x80000) {
986        /* Combine the low 20 bits of the offset with the actual load insn;
987           the high 44 bits must come from an immediate load.  */
988        tcg_target_long low = ((ofs & 0xfffff) ^ 0x80000) - 0x80000;
989        tcg_out_movi(s, TCG_TYPE_PTR, TCG_TMP0, ofs - low);
990        ofs = low;
991
992        /* If we were already given an index register, add it in.  */
993        if (index != TCG_REG_NONE) {
994            tcg_out_insn(s, RRE, AGR, TCG_TMP0, index);
995        }
996        index = TCG_TMP0;
997    }
998
999    if (opc_rx && ofs >= 0 && ofs < 0x1000) {
1000        tcg_out_insn_RX(s, opc_rx, data, base, index, ofs);
1001    } else {
1002        tcg_out_insn_RXY(s, opc_rxy, data, base, index, ofs);
1003    }
1004}
1005
1006static void tcg_out_vrx_mem(TCGContext *s, S390Opcode opc_vrx,
1007                            TCGReg data, TCGReg base, TCGReg index,
1008                            tcg_target_long ofs, int m3)
1009{
1010    if (ofs < 0 || ofs >= 0x1000) {
1011        if (ofs >= -0x80000 && ofs < 0x80000) {
1012            tcg_out_insn(s, RXY, LAY, TCG_TMP0, base, index, ofs);
1013            base = TCG_TMP0;
1014            index = TCG_REG_NONE;
1015            ofs = 0;
1016        } else {
1017            tcg_out_movi(s, TCG_TYPE_PTR, TCG_TMP0, ofs);
1018            if (index != TCG_REG_NONE) {
1019                tcg_out_insn(s, RRE, AGR, TCG_TMP0, index);
1020            }
1021            index = TCG_TMP0;
1022            ofs = 0;
1023        }
1024    }
1025    tcg_out_insn_VRX(s, opc_vrx, data, base, index, ofs, m3);
1026}
1027
1028/* load data without address translation or endianness conversion */
1029static void tcg_out_ld(TCGContext *s, TCGType type, TCGReg data,
1030                       TCGReg base, intptr_t ofs)
1031{
1032    switch (type) {
1033    case TCG_TYPE_I32:
1034        if (likely(is_general_reg(data))) {
1035            tcg_out_mem(s, RX_L, RXY_LY, data, base, TCG_REG_NONE, ofs);
1036            break;
1037        }
1038        tcg_out_vrx_mem(s, VRX_VLLEZ, data, base, TCG_REG_NONE, ofs, MO_32);
1039        break;
1040
1041    case TCG_TYPE_I64:
1042        if (likely(is_general_reg(data))) {
1043            tcg_out_mem(s, 0, RXY_LG, data, base, TCG_REG_NONE, ofs);
1044            break;
1045        }
1046        /* fallthru */
1047
1048    case TCG_TYPE_V64:
1049        tcg_out_vrx_mem(s, VRX_VLLEZ, data, base, TCG_REG_NONE, ofs, MO_64);
1050        break;
1051
1052    case TCG_TYPE_V128:
1053        /* Hint quadword aligned.  */
1054        tcg_out_vrx_mem(s, VRX_VL, data, base, TCG_REG_NONE, ofs, 4);
1055        break;
1056
1057    default:
1058        g_assert_not_reached();
1059    }
1060}
1061
1062static void tcg_out_st(TCGContext *s, TCGType type, TCGReg data,
1063                       TCGReg base, intptr_t ofs)
1064{
1065    switch (type) {
1066    case TCG_TYPE_I32:
1067        if (likely(is_general_reg(data))) {
1068            tcg_out_mem(s, RX_ST, RXY_STY, data, base, TCG_REG_NONE, ofs);
1069        } else {
1070            tcg_out_vrx_mem(s, VRX_VSTEF, data, base, TCG_REG_NONE, ofs, 1);
1071        }
1072        break;
1073
1074    case TCG_TYPE_I64:
1075        if (likely(is_general_reg(data))) {
1076            tcg_out_mem(s, 0, RXY_STG, data, base, TCG_REG_NONE, ofs);
1077            break;
1078        }
1079        /* fallthru */
1080
1081    case TCG_TYPE_V64:
1082        tcg_out_vrx_mem(s, VRX_VSTEG, data, base, TCG_REG_NONE, ofs, 0);
1083        break;
1084
1085    case TCG_TYPE_V128:
1086        /* Hint quadword aligned.  */
1087        tcg_out_vrx_mem(s, VRX_VST, data, base, TCG_REG_NONE, ofs, 4);
1088        break;
1089
1090    default:
1091        g_assert_not_reached();
1092    }
1093}
1094
1095static inline bool tcg_out_sti(TCGContext *s, TCGType type, TCGArg val,
1096                               TCGReg base, intptr_t ofs)
1097{
1098    return false;
1099}
1100
1101static bool tcg_out_xchg(TCGContext *s, TCGType type, TCGReg r1, TCGReg r2)
1102{
1103    return false;
1104}
1105
1106static void tcg_out_addi_ptr(TCGContext *s, TCGReg rd, TCGReg rs,
1107                             tcg_target_long imm)
1108{
1109    /* This function is only used for passing structs by reference. */
1110    tcg_out_mem(s, RX_LA, RXY_LAY, rd, rs, TCG_REG_NONE, imm);
1111}
1112
1113static inline void tcg_out_risbg(TCGContext *s, TCGReg dest, TCGReg src,
1114                                 int msb, int lsb, int ofs, int z)
1115{
1116    /* Format RIE-f */
1117    tcg_out16(s, (RIEf_RISBG & 0xff00) | (dest << 4) | src);
1118    tcg_out16(s, (msb << 8) | (z << 7) | lsb);
1119    tcg_out16(s, (ofs << 8) | (RIEf_RISBG & 0xff));
1120}
1121
1122static void tcg_out_ext8s(TCGContext *s, TCGType type, TCGReg dest, TCGReg src)
1123{
1124    tcg_out_insn(s, RRE, LGBR, dest, src);
1125}
1126
1127static void tcg_out_ext8u(TCGContext *s, TCGReg dest, TCGReg src)
1128{
1129    tcg_out_insn(s, RRE, LLGCR, dest, src);
1130}
1131
1132static void tcg_out_ext16s(TCGContext *s, TCGType type, TCGReg dest, TCGReg src)
1133{
1134    tcg_out_insn(s, RRE, LGHR, dest, src);
1135}
1136
1137static void tcg_out_ext16u(TCGContext *s, TCGReg dest, TCGReg src)
1138{
1139    tcg_out_insn(s, RRE, LLGHR, dest, src);
1140}
1141
1142static void tcg_out_ext32s(TCGContext *s, TCGReg dest, TCGReg src)
1143{
1144    tcg_out_insn(s, RRE, LGFR, dest, src);
1145}
1146
1147static void tcg_out_ext32u(TCGContext *s, TCGReg dest, TCGReg src)
1148{
1149    tcg_out_insn(s, RRE, LLGFR, dest, src);
1150}
1151
1152static void tcg_out_exts_i32_i64(TCGContext *s, TCGReg dest, TCGReg src)
1153{
1154    tcg_out_ext32s(s, dest, src);
1155}
1156
1157static void tcg_out_extu_i32_i64(TCGContext *s, TCGReg dest, TCGReg src)
1158{
1159    tcg_out_ext32u(s, dest, src);
1160}
1161
1162static void tcg_out_extrl_i64_i32(TCGContext *s, TCGReg dest, TCGReg src)
1163{
1164    tcg_out_mov(s, TCG_TYPE_I32, dest, src);
1165}
1166
1167static void tgen_andi_risbg(TCGContext *s, TCGReg out, TCGReg in, uint64_t val)
1168{
1169    int msb, lsb;
1170    if ((val & 0x8000000000000001ull) == 0x8000000000000001ull) {
1171        /* Achieve wraparound by swapping msb and lsb.  */
1172        msb = 64 - ctz64(~val);
1173        lsb = clz64(~val) - 1;
1174    } else {
1175        msb = clz64(val);
1176        lsb = 63 - ctz64(val);
1177    }
1178    tcg_out_risbg(s, out, in, msb, lsb, 0, 1);
1179}
1180
1181static void tgen_andi(TCGContext *s, TCGType type, TCGReg dest, uint64_t val)
1182{
1183    static const S390Opcode ni_insns[4] = {
1184        RI_NILL, RI_NILH, RI_NIHL, RI_NIHH
1185    };
1186    static const S390Opcode nif_insns[2] = {
1187        RIL_NILF, RIL_NIHF
1188    };
1189    uint64_t valid = (type == TCG_TYPE_I32 ? 0xffffffffull : -1ull);
1190    int i;
1191
1192    /* Look for the zero-extensions.  */
1193    if ((val & valid) == 0xffffffff) {
1194        tcg_out_ext32u(s, dest, dest);
1195        return;
1196    }
1197    if ((val & valid) == 0xff) {
1198        tcg_out_ext8u(s, dest, dest);
1199        return;
1200    }
1201    if ((val & valid) == 0xffff) {
1202        tcg_out_ext16u(s, dest, dest);
1203        return;
1204    }
1205
1206    i = is_const_p16(~val & valid);
1207    if (i >= 0) {
1208        tcg_out_insn_RI(s, ni_insns[i], dest, val >> (i * 16));
1209        return;
1210    }
1211
1212    i = is_const_p32(~val & valid);
1213    tcg_debug_assert(i == 0 || type != TCG_TYPE_I32);
1214    if (i >= 0) {
1215        tcg_out_insn_RIL(s, nif_insns[i], dest, val >> (i * 32));
1216        return;
1217    }
1218
1219    if (risbg_mask(val)) {
1220        tgen_andi_risbg(s, dest, dest, val);
1221        return;
1222    }
1223
1224    g_assert_not_reached();
1225}
1226
1227static void tgen_ori(TCGContext *s, TCGReg dest, uint64_t val)
1228{
1229    static const S390Opcode oif_insns[2] = {
1230        RIL_OILF, RIL_OIHF
1231    };
1232
1233    int i;
1234
1235    i = is_const_p16(val);
1236    if (i >= 0) {
1237        tcg_out_insn_RI(s, oi_insns[i], dest, val >> (i * 16));
1238        return;
1239    }
1240
1241    i = is_const_p32(val);
1242    if (i >= 0) {
1243        tcg_out_insn_RIL(s, oif_insns[i], dest, val >> (i * 32));
1244        return;
1245    }
1246
1247    g_assert_not_reached();
1248}
1249
1250static void tgen_xori(TCGContext *s, TCGReg dest, uint64_t val)
1251{
1252    switch (is_const_p32(val)) {
1253    case 0:
1254        tcg_out_insn(s, RIL, XILF, dest, val);
1255        break;
1256    case 1:
1257        tcg_out_insn(s, RIL, XIHF, dest, val >> 32);
1258        break;
1259    default:
1260        g_assert_not_reached();
1261    }
1262}
1263
1264static int tgen_cmp2(TCGContext *s, TCGType type, TCGCond c, TCGReg r1,
1265                     TCGArg c2, bool c2const, bool need_carry, int *inv_cc)
1266{
1267    bool is_unsigned = is_unsigned_cond(c);
1268    TCGCond inv_c = tcg_invert_cond(c);
1269    S390Opcode op;
1270
1271    if (is_tst_cond(c)) {
1272        tcg_debug_assert(!need_carry);
1273
1274        if (!c2const) {
1275            if (type == TCG_TYPE_I32) {
1276                tcg_out_insn(s, RRFa, NRK, TCG_REG_R0, r1, c2);
1277            } else {
1278                tcg_out_insn(s, RRFa, NGRK, TCG_REG_R0, r1, c2);
1279            }
1280            goto exit;
1281        }
1282
1283        if (type == TCG_TYPE_I32) {
1284            c2 = (uint32_t)c2;
1285        }
1286
1287        int i = is_const_p16(c2);
1288        if (i >= 0) {
1289            tcg_out_insn_RI(s, tm_insns[i], r1, c2 >> (i * 16));
1290            *inv_cc = c == TCG_COND_TSTEQ ? S390_TM_NE : S390_TM_EQ;
1291            return *inv_cc ^ 15;
1292        }
1293
1294        if (risbg_mask(c2)) {
1295            tgen_andi_risbg(s, TCG_REG_R0, r1, c2);
1296            goto exit;
1297        }
1298        g_assert_not_reached();
1299    }
1300
1301    if (c2const) {
1302        if (c2 == 0) {
1303            if (!(is_unsigned && need_carry)) {
1304                if (type == TCG_TYPE_I32) {
1305                    tcg_out_insn(s, RR, LTR, r1, r1);
1306                } else {
1307                    tcg_out_insn(s, RRE, LTGR, r1, r1);
1308                }
1309                *inv_cc = tcg_cond_to_ltr_cond[inv_c];
1310                return tcg_cond_to_ltr_cond[c];
1311            }
1312        }
1313
1314        if (!is_unsigned && c2 == (int16_t)c2) {
1315            op = (type == TCG_TYPE_I32 ? RI_CHI : RI_CGHI);
1316            tcg_out_insn_RI(s, op, r1, c2);
1317            goto exit;
1318        }
1319
1320        if (type == TCG_TYPE_I32) {
1321            op = (is_unsigned ? RIL_CLFI : RIL_CFI);
1322            tcg_out_insn_RIL(s, op, r1, c2);
1323            goto exit;
1324        }
1325
1326        /* Should match TCG_CT_CONST_CMP. */
1327        switch (c) {
1328        case TCG_COND_LT:
1329        case TCG_COND_GE:
1330        case TCG_COND_LE:
1331        case TCG_COND_GT:
1332            tcg_debug_assert(c2 == (int32_t)c2);
1333            op = RIL_CGFI;
1334            break;
1335        case TCG_COND_EQ:
1336        case TCG_COND_NE:
1337            if (c2 == (int32_t)c2) {
1338                op = RIL_CGFI;
1339                break;
1340            }
1341            /* fall through */
1342        case TCG_COND_LTU:
1343        case TCG_COND_GEU:
1344        case TCG_COND_LEU:
1345        case TCG_COND_GTU:
1346            tcg_debug_assert(c2 == (uint32_t)c2);
1347            op = RIL_CLGFI;
1348            break;
1349        default:
1350            g_assert_not_reached();
1351        }
1352        tcg_out_insn_RIL(s, op, r1, c2);
1353    } else if (type == TCG_TYPE_I32) {
1354        op = (is_unsigned ? RR_CLR : RR_CR);
1355        tcg_out_insn_RR(s, op, r1, c2);
1356    } else {
1357        op = (is_unsigned ? RRE_CLGR : RRE_CGR);
1358        tcg_out_insn_RRE(s, op, r1, c2);
1359    }
1360
1361 exit:
1362    *inv_cc = tcg_cond_to_s390_cond[inv_c];
1363    return tcg_cond_to_s390_cond[c];
1364}
1365
1366static int tgen_cmp(TCGContext *s, TCGType type, TCGCond c, TCGReg r1,
1367                    TCGArg c2, bool c2const, bool need_carry)
1368{
1369    int inv_cc;
1370    return tgen_cmp2(s, type, c, r1, c2, c2const, need_carry, &inv_cc);
1371}
1372
1373static void tgen_setcond_int(TCGContext *s, TCGType type, TCGCond cond,
1374                             TCGReg dest, TCGReg c1, TCGArg c2,
1375                             bool c2const, bool neg)
1376{
1377    int cc;
1378
1379    /* With LOC2, we can always emit the minimum 3 insns.  */
1380    if (HAVE_FACILITY(LOAD_ON_COND2)) {
1381        /* Emit: d = 0, d = (cc ? 1 : d).  */
1382        cc = tgen_cmp(s, type, cond, c1, c2, c2const, false);
1383        tcg_out_movi(s, TCG_TYPE_I64, dest, 0);
1384        tcg_out_insn(s, RIEg, LOCGHI, dest, neg ? -1 : 1, cc);
1385        return;
1386    }
1387
1388    switch (cond) {
1389    case TCG_COND_GEU:
1390    case TCG_COND_LTU:
1391    case TCG_COND_LT:
1392    case TCG_COND_GE:
1393        /* Swap operands so that we can use LEU/GTU/GT/LE.  */
1394        if (!c2const) {
1395            TCGReg t = c1;
1396            c1 = c2;
1397            c2 = t;
1398            cond = tcg_swap_cond(cond);
1399        }
1400        break;
1401    default:
1402        break;
1403    }
1404
1405    switch (cond) {
1406    case TCG_COND_NE:
1407        /* X != 0 is X > 0.  */
1408        if (c2const && c2 == 0) {
1409            cond = TCG_COND_GTU;
1410        } else {
1411            break;
1412        }
1413        /* fallthru */
1414
1415    case TCG_COND_GTU:
1416    case TCG_COND_GT:
1417        /*
1418         * The result of a compare has CC=2 for GT and CC=3 unused.
1419         * ADD LOGICAL WITH CARRY considers (CC & 2) the carry bit.
1420         */
1421        tgen_cmp(s, type, cond, c1, c2, c2const, true);
1422        tcg_out_movi(s, type, dest, 0);
1423        tcg_out_insn(s, RRE, ALCGR, dest, dest);
1424        if (neg) {
1425            if (type == TCG_TYPE_I32) {
1426                tcg_out_insn(s, RR, LCR, dest, dest);
1427            } else {
1428                tcg_out_insn(s, RRE, LCGR, dest, dest);
1429            }
1430        }
1431        return;
1432
1433    case TCG_COND_EQ:
1434        /* X == 0 is X <= 0.  */
1435        if (c2const && c2 == 0) {
1436            cond = TCG_COND_LEU;
1437        } else {
1438            break;
1439        }
1440        /* fallthru */
1441
1442    case TCG_COND_LEU:
1443    case TCG_COND_LE:
1444        /*
1445         * As above, but we're looking for borrow, or !carry.
1446         * The second insn computes d - d - borrow, or -1 for true
1447         * and 0 for false.  So we must mask to 1 bit afterward.
1448         */
1449        tgen_cmp(s, type, cond, c1, c2, c2const, true);
1450        tcg_out_insn(s, RRE, SLBGR, dest, dest);
1451        if (!neg) {
1452            tgen_andi(s, type, dest, 1);
1453        }
1454        return;
1455
1456    default:
1457        g_assert_not_reached();
1458    }
1459
1460    cc = tgen_cmp(s, type, cond, c1, c2, c2const, false);
1461    /* Emit: d = 0, t = 1, d = (cc ? t : d).  */
1462    tcg_out_movi(s, TCG_TYPE_I64, dest, 0);
1463    tcg_out_movi(s, TCG_TYPE_I64, TCG_TMP0, neg ? -1 : 1);
1464    tcg_out_insn(s, RRFc, LOCGR, dest, TCG_TMP0, cc);
1465}
1466
1467static void tgen_setcond(TCGContext *s, TCGType type, TCGCond cond,
1468                         TCGReg dest, TCGReg arg1, TCGReg arg2)
1469{
1470    tgen_setcond_int(s, type, cond, dest, arg1, arg2, false, false);
1471}
1472
1473static void tgen_setcondi(TCGContext *s, TCGType type, TCGCond cond,
1474                          TCGReg dest, TCGReg arg1, tcg_target_long arg2)
1475{
1476    tgen_setcond_int(s, type, cond, dest, arg1, arg2, true, false);
1477}
1478
1479static const TCGOutOpSetcond outop_setcond = {
1480    .base.static_constraint = C_O1_I2(r, r, rC),
1481    .out_rrr = tgen_setcond,
1482    .out_rri = tgen_setcondi,
1483};
1484
1485static void tgen_negsetcond(TCGContext *s, TCGType type, TCGCond cond,
1486                            TCGReg dest, TCGReg arg1, TCGReg arg2)
1487{
1488    tgen_setcond_int(s, type, cond, dest, arg1, arg2, false, true);
1489}
1490
1491static void tgen_negsetcondi(TCGContext *s, TCGType type, TCGCond cond,
1492                             TCGReg dest, TCGReg arg1, tcg_target_long arg2)
1493{
1494    tgen_setcond_int(s, type, cond, dest, arg1, arg2, true, true);
1495}
1496
1497static const TCGOutOpSetcond outop_negsetcond = {
1498    .base.static_constraint = C_O1_I2(r, r, rC),
1499    .out_rrr = tgen_negsetcond,
1500    .out_rri = tgen_negsetcondi,
1501};
1502
1503static void tgen_movcond_int(TCGContext *s, TCGType type, TCGReg dest,
1504                             TCGArg v3, int v3const, TCGReg v4,
1505                             int cc, int inv_cc)
1506{
1507    TCGReg src;
1508
1509    if (v3const) {
1510        if (dest == v4) {
1511            if (HAVE_FACILITY(LOAD_ON_COND2)) {
1512                /* Emit: if (cc) dest = v3. */
1513                tcg_out_insn(s, RIEg, LOCGHI, dest, v3, cc);
1514                return;
1515            }
1516            tcg_out_insn(s, RI, LGHI, TCG_TMP0, v3);
1517            src = TCG_TMP0;
1518        } else {
1519            /* LGR+LOCGHI is larger than LGHI+LOCGR. */
1520            tcg_out_insn(s, RI, LGHI, dest, v3);
1521            cc = inv_cc;
1522            src = v4;
1523        }
1524    } else {
1525        if (HAVE_FACILITY(MISC_INSN_EXT3)) {
1526            /* Emit: dest = cc ? v3 : v4. */
1527            tcg_out_insn(s, RRFam, SELGR, dest, v3, v4, cc);
1528            return;
1529        }
1530        if (dest == v4) {
1531            src = v3;
1532        } else {
1533            tcg_out_mov(s, type, dest, v3);
1534            cc = inv_cc;
1535            src = v4;
1536        }
1537    }
1538
1539    /* Emit: if (cc) dest = src. */
1540    tcg_out_insn(s, RRFc, LOCGR, dest, src, cc);
1541}
1542
1543static void tgen_movcond(TCGContext *s, TCGType type, TCGCond c,
1544                         TCGReg dest, TCGReg c1, TCGArg c2, bool c2const,
1545                         TCGArg v3, bool v3const, TCGArg v4, bool v4const)
1546{
1547    int cc, inv_cc;
1548
1549    cc = tgen_cmp2(s, type, c, c1, c2, c2const, false, &inv_cc);
1550    tgen_movcond_int(s, type, dest, v3, v3const, v4, cc, inv_cc);
1551}
1552
1553static const TCGOutOpMovcond outop_movcond = {
1554    .base.static_constraint = C_O1_I4(r, r, rC, rI, r),
1555    .out = tgen_movcond,
1556};
1557
1558static void tgen_deposit(TCGContext *s, TCGReg dest, TCGReg src,
1559                         int ofs, int len, int z)
1560{
1561    int lsb = (63 - ofs);
1562    int msb = lsb - (len - 1);
1563    tcg_out_risbg(s, dest, src, msb, lsb, ofs, z);
1564}
1565
1566static void tgen_extract(TCGContext *s, TCGType type, TCGReg dest,
1567                         TCGReg src, unsigned ofs, unsigned len)
1568{
1569    if (ofs == 0) {
1570        switch (len) {
1571        case 8:
1572            tcg_out_ext8u(s, dest, src);
1573            return;
1574        case 16:
1575            tcg_out_ext16u(s, dest, src);
1576            return;
1577        case 32:
1578            tcg_out_ext32u(s, dest, src);
1579            return;
1580        }
1581    }
1582    tcg_out_risbg(s, dest, src, 64 - len, 63, 64 - ofs, 1);
1583}
1584
1585static const TCGOutOpExtract outop_extract = {
1586    .base.static_constraint = C_O1_I1(r, r),
1587    .out_rr = tgen_extract,
1588};
1589
1590static void tgen_sextract(TCGContext *s, TCGReg dest, TCGReg src,
1591                          int ofs, int len)
1592{
1593    if (ofs == 0) {
1594        switch (len) {
1595        case 8:
1596            tcg_out_ext8s(s, TCG_TYPE_REG, dest, src);
1597            return;
1598        case 16:
1599            tcg_out_ext16s(s, TCG_TYPE_REG, dest, src);
1600            return;
1601        case 32:
1602            tcg_out_ext32s(s, dest, src);
1603            return;
1604        }
1605    }
1606    g_assert_not_reached();
1607}
1608
1609static void tgen_gotoi(TCGContext *s, int cc, const tcg_insn_unit *dest)
1610{
1611    ptrdiff_t off = tcg_pcrel_diff(s, dest) >> 1;
1612    if (off == (int16_t)off) {
1613        tcg_out_insn(s, RI, BRC, cc, off);
1614    } else if (off == (int32_t)off) {
1615        tcg_out_insn(s, RIL, BRCL, cc, off);
1616    } else {
1617        tcg_out_movi(s, TCG_TYPE_PTR, TCG_TMP0, (uintptr_t)dest);
1618        tcg_out_insn(s, RR, BCR, cc, TCG_TMP0);
1619    }
1620}
1621
1622static void tgen_branch(TCGContext *s, int cc, TCGLabel *l)
1623{
1624    if (l->has_value) {
1625        tgen_gotoi(s, cc, l->u.value_ptr);
1626    } else {
1627        tcg_out16(s, RI_BRC | (cc << 4));
1628        tcg_out_reloc(s, s->code_ptr, R_390_PC16DBL, l, 2);
1629        s->code_ptr += 1;
1630    }
1631}
1632
1633static void tgen_compare_branch(TCGContext *s, S390Opcode opc, int cc,
1634                                TCGReg r1, TCGReg r2, TCGLabel *l)
1635{
1636    tcg_out_reloc(s, s->code_ptr + 1, R_390_PC16DBL, l, 2);
1637    /* Format RIE-b */
1638    tcg_out16(s, (opc & 0xff00) | (r1 << 4) | r2);
1639    tcg_out16(s, 0);
1640    tcg_out16(s, cc << 12 | (opc & 0xff));
1641}
1642
1643static void tgen_compare_imm_branch(TCGContext *s, S390Opcode opc, int cc,
1644                                    TCGReg r1, int i2, TCGLabel *l)
1645{
1646    tcg_out_reloc(s, s->code_ptr + 1, R_390_PC16DBL, l, 2);
1647    /* Format RIE-c */
1648    tcg_out16(s, (opc & 0xff00) | (r1 << 4) | cc);
1649    tcg_out16(s, 0);
1650    tcg_out16(s, (i2 << 8) | (opc & 0xff));
1651}
1652
1653static void tgen_brcond(TCGContext *s, TCGType type, TCGCond c,
1654                        TCGReg r1, TCGArg c2, int c2const, TCGLabel *l)
1655{
1656    int cc;
1657
1658    if (!is_tst_cond(c)) {
1659        bool is_unsigned = is_unsigned_cond(c);
1660        bool in_range;
1661        S390Opcode opc;
1662
1663        cc = tcg_cond_to_s390_cond[c];
1664
1665        if (!c2const) {
1666            opc = (type == TCG_TYPE_I32
1667                   ? (is_unsigned ? RIEb_CLRJ : RIEb_CRJ)
1668                   : (is_unsigned ? RIEb_CLGRJ : RIEb_CGRJ));
1669            tgen_compare_branch(s, opc, cc, r1, c2, l);
1670            return;
1671        }
1672
1673        /*
1674         * COMPARE IMMEDIATE AND BRANCH RELATIVE has an 8-bit immediate field.
1675         * If the immediate we've been given does not fit that range, we'll
1676         * fall back to separate compare and branch instructions using the
1677         * larger comparison range afforded by COMPARE IMMEDIATE.
1678         */
1679        if (type == TCG_TYPE_I32) {
1680            if (is_unsigned) {
1681                opc = RIEc_CLIJ;
1682                in_range = (uint32_t)c2 == (uint8_t)c2;
1683            } else {
1684                opc = RIEc_CIJ;
1685                in_range = (int32_t)c2 == (int8_t)c2;
1686            }
1687        } else {
1688            if (is_unsigned) {
1689                opc = RIEc_CLGIJ;
1690                in_range = (uint64_t)c2 == (uint8_t)c2;
1691            } else {
1692                opc = RIEc_CGIJ;
1693                in_range = (int64_t)c2 == (int8_t)c2;
1694            }
1695        }
1696        if (in_range) {
1697            tgen_compare_imm_branch(s, opc, cc, r1, c2, l);
1698            return;
1699        }
1700    }
1701
1702    cc = tgen_cmp(s, type, c, r1, c2, c2const, false);
1703    tgen_branch(s, cc, l);
1704}
1705
1706static void tgen_brcondr(TCGContext *s, TCGType type, TCGCond c,
1707                         TCGReg a0, TCGReg a1, TCGLabel *l)
1708{
1709    tgen_brcond(s, type, c, a0, a1, false, l);
1710}
1711
1712static void tgen_brcondi(TCGContext *s, TCGType type, TCGCond c,
1713                         TCGReg a0, tcg_target_long a1, TCGLabel *l)
1714{
1715    tgen_brcond(s, type, c, a0, a1, true, l);
1716}
1717
1718static const TCGOutOpBrcond outop_brcond = {
1719    .base.static_constraint = C_O0_I2(r, rC),
1720    .out_rr = tgen_brcondr,
1721    .out_ri = tgen_brcondi,
1722};
1723
1724static void tcg_out_call_int(TCGContext *s, const tcg_insn_unit *dest)
1725{
1726    ptrdiff_t off = tcg_pcrel_diff(s, dest) >> 1;
1727    if (off == (int32_t)off) {
1728        tcg_out_insn(s, RIL, BRASL, TCG_REG_R14, off);
1729    } else {
1730        tcg_out_movi(s, TCG_TYPE_PTR, TCG_TMP0, (uintptr_t)dest);
1731        tcg_out_insn(s, RR, BASR, TCG_REG_R14, TCG_TMP0);
1732    }
1733}
1734
1735static void tcg_out_call(TCGContext *s, const tcg_insn_unit *dest,
1736                         const TCGHelperInfo *info)
1737{
1738    tcg_out_call_int(s, dest);
1739}
1740
1741typedef struct {
1742    TCGReg base;
1743    TCGReg index;
1744    int disp;
1745    TCGAtomAlign aa;
1746} HostAddress;
1747
1748bool tcg_target_has_memory_bswap(MemOp memop)
1749{
1750    TCGAtomAlign aa;
1751
1752    if ((memop & MO_SIZE) <= MO_64) {
1753        return true;
1754    }
1755
1756    /*
1757     * Reject 16-byte memop with 16-byte atomicity,
1758     * but do allow a pair of 64-bit operations.
1759     */
1760    aa = atom_and_align_for_opc(tcg_ctx, memop, MO_ATOM_IFALIGN, true);
1761    return aa.atom <= MO_64;
1762}
1763
1764static void tcg_out_qemu_ld_direct(TCGContext *s, MemOp opc, TCGReg data,
1765                                   HostAddress h)
1766{
1767    switch (opc & (MO_SSIZE | MO_BSWAP)) {
1768    case MO_UB:
1769        tcg_out_insn(s, RXY, LLGC, data, h.base, h.index, h.disp);
1770        break;
1771    case MO_SB:
1772        tcg_out_insn(s, RXY, LGB, data, h.base, h.index, h.disp);
1773        break;
1774
1775    case MO_UW | MO_BSWAP:
1776        /* swapped unsigned halfword load with upper bits zeroed */
1777        tcg_out_insn(s, RXY, LRVH, data, h.base, h.index, h.disp);
1778        tcg_out_ext16u(s, data, data);
1779        break;
1780    case MO_UW:
1781        tcg_out_insn(s, RXY, LLGH, data, h.base, h.index, h.disp);
1782        break;
1783
1784    case MO_SW | MO_BSWAP:
1785        /* swapped sign-extended halfword load */
1786        tcg_out_insn(s, RXY, LRVH, data, h.base, h.index, h.disp);
1787        tcg_out_ext16s(s, TCG_TYPE_REG, data, data);
1788        break;
1789    case MO_SW:
1790        tcg_out_insn(s, RXY, LGH, data, h.base, h.index, h.disp);
1791        break;
1792
1793    case MO_UL | MO_BSWAP:
1794        /* swapped unsigned int load with upper bits zeroed */
1795        tcg_out_insn(s, RXY, LRV, data, h.base, h.index, h.disp);
1796        tcg_out_ext32u(s, data, data);
1797        break;
1798    case MO_UL:
1799        tcg_out_insn(s, RXY, LLGF, data, h.base, h.index, h.disp);
1800        break;
1801
1802    case MO_SL | MO_BSWAP:
1803        /* swapped sign-extended int load */
1804        tcg_out_insn(s, RXY, LRV, data, h.base, h.index, h.disp);
1805        tcg_out_ext32s(s, data, data);
1806        break;
1807    case MO_SL:
1808        tcg_out_insn(s, RXY, LGF, data, h.base, h.index, h.disp);
1809        break;
1810
1811    case MO_UQ | MO_BSWAP:
1812        tcg_out_insn(s, RXY, LRVG, data, h.base, h.index, h.disp);
1813        break;
1814    case MO_UQ:
1815        tcg_out_insn(s, RXY, LG, data, h.base, h.index, h.disp);
1816        break;
1817
1818    default:
1819        g_assert_not_reached();
1820    }
1821}
1822
1823static void tcg_out_qemu_st_direct(TCGContext *s, MemOp opc, TCGReg data,
1824                                   HostAddress h)
1825{
1826    switch (opc & (MO_SIZE | MO_BSWAP)) {
1827    case MO_UB:
1828        if (h.disp >= 0 && h.disp < 0x1000) {
1829            tcg_out_insn(s, RX, STC, data, h.base, h.index, h.disp);
1830        } else {
1831            tcg_out_insn(s, RXY, STCY, data, h.base, h.index, h.disp);
1832        }
1833        break;
1834
1835    case MO_UW | MO_BSWAP:
1836        tcg_out_insn(s, RXY, STRVH, data, h.base, h.index, h.disp);
1837        break;
1838    case MO_UW:
1839        if (h.disp >= 0 && h.disp < 0x1000) {
1840            tcg_out_insn(s, RX, STH, data, h.base, h.index, h.disp);
1841        } else {
1842            tcg_out_insn(s, RXY, STHY, data, h.base, h.index, h.disp);
1843        }
1844        break;
1845
1846    case MO_UL | MO_BSWAP:
1847        tcg_out_insn(s, RXY, STRV, data, h.base, h.index, h.disp);
1848        break;
1849    case MO_UL:
1850        if (h.disp >= 0 && h.disp < 0x1000) {
1851            tcg_out_insn(s, RX, ST, data, h.base, h.index, h.disp);
1852        } else {
1853            tcg_out_insn(s, RXY, STY, data, h.base, h.index, h.disp);
1854        }
1855        break;
1856
1857    case MO_UQ | MO_BSWAP:
1858        tcg_out_insn(s, RXY, STRVG, data, h.base, h.index, h.disp);
1859        break;
1860    case MO_UQ:
1861        tcg_out_insn(s, RXY, STG, data, h.base, h.index, h.disp);
1862        break;
1863
1864    default:
1865        g_assert_not_reached();
1866    }
1867}
1868
1869static const TCGLdstHelperParam ldst_helper_param = {
1870    .ntmp = 1, .tmp = { TCG_TMP0 }
1871};
1872
1873static bool tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
1874{
1875    MemOp opc = get_memop(lb->oi);
1876
1877    if (!patch_reloc(lb->label_ptr[0], R_390_PC16DBL,
1878                     (intptr_t)tcg_splitwx_to_rx(s->code_ptr), 2)) {
1879        return false;
1880    }
1881
1882    tcg_out_ld_helper_args(s, lb, &ldst_helper_param);
1883    tcg_out_call_int(s, qemu_ld_helpers[opc & MO_SIZE]);
1884    tcg_out_ld_helper_ret(s, lb, false, &ldst_helper_param);
1885
1886    tgen_gotoi(s, S390_CC_ALWAYS, lb->raddr);
1887    return true;
1888}
1889
1890static bool tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
1891{
1892    MemOp opc = get_memop(lb->oi);
1893
1894    if (!patch_reloc(lb->label_ptr[0], R_390_PC16DBL,
1895                     (intptr_t)tcg_splitwx_to_rx(s->code_ptr), 2)) {
1896        return false;
1897    }
1898
1899    tcg_out_st_helper_args(s, lb, &ldst_helper_param);
1900    tcg_out_call_int(s, qemu_st_helpers[opc & MO_SIZE]);
1901
1902    tgen_gotoi(s, S390_CC_ALWAYS, lb->raddr);
1903    return true;
1904}
1905
1906/* We're expecting to use a 20-bit negative offset on the tlb memory ops.  */
1907#define MIN_TLB_MASK_TABLE_OFS  -(1 << 19)
1908
1909/*
1910 * For system-mode, perform the TLB load and compare.
1911 * For user-mode, perform any required alignment tests.
1912 * In both cases, return a TCGLabelQemuLdst structure if the slow path
1913 * is required and fill in @h with the host address for the fast path.
1914 */
1915static TCGLabelQemuLdst *prepare_host_addr(TCGContext *s, HostAddress *h,
1916                                           TCGReg addr_reg, MemOpIdx oi,
1917                                           bool is_ld)
1918{
1919    TCGType addr_type = s->addr_type;
1920    TCGLabelQemuLdst *ldst = NULL;
1921    MemOp opc = get_memop(oi);
1922    MemOp s_bits = opc & MO_SIZE;
1923    unsigned a_mask;
1924
1925    h->aa = atom_and_align_for_opc(s, opc, MO_ATOM_IFALIGN, s_bits == MO_128);
1926    a_mask = (1 << h->aa.align) - 1;
1927
1928    if (tcg_use_softmmu) {
1929        unsigned s_mask = (1 << s_bits) - 1;
1930        int mem_index = get_mmuidx(oi);
1931        int fast_off = tlb_mask_table_ofs(s, mem_index);
1932        int mask_off = fast_off + offsetof(CPUTLBDescFast, mask);
1933        int table_off = fast_off + offsetof(CPUTLBDescFast, table);
1934        int ofs, a_off;
1935        uint64_t tlb_mask;
1936
1937        ldst = new_ldst_label(s);
1938        ldst->is_ld = is_ld;
1939        ldst->oi = oi;
1940        ldst->addr_reg = addr_reg;
1941
1942        tcg_out_sh64(s, RSY_SRLG, TCG_TMP0, addr_reg, TCG_REG_NONE,
1943                     s->page_bits - CPU_TLB_ENTRY_BITS);
1944
1945        tcg_out_insn(s, RXY, NG, TCG_TMP0, TCG_AREG0, TCG_REG_NONE, mask_off);
1946        tcg_out_insn(s, RXY, AG, TCG_TMP0, TCG_AREG0, TCG_REG_NONE, table_off);
1947
1948        /*
1949         * For aligned accesses, we check the first byte and include the
1950         * alignment bits within the address.  For unaligned access, we
1951         * check that we don't cross pages using the address of the last
1952         * byte of the access.
1953         */
1954        a_off = (a_mask >= s_mask ? 0 : s_mask - a_mask);
1955        tlb_mask = (uint64_t)s->page_mask | a_mask;
1956        if (a_off == 0) {
1957            tgen_andi_risbg(s, TCG_REG_R0, addr_reg, tlb_mask);
1958        } else {
1959            tcg_out_insn(s, RX, LA, TCG_REG_R0, addr_reg, TCG_REG_NONE, a_off);
1960            tgen_andi(s, addr_type, TCG_REG_R0, tlb_mask);
1961        }
1962
1963        if (is_ld) {
1964            ofs = offsetof(CPUTLBEntry, addr_read);
1965        } else {
1966            ofs = offsetof(CPUTLBEntry, addr_write);
1967        }
1968        if (addr_type == TCG_TYPE_I32) {
1969            ofs += HOST_BIG_ENDIAN * 4;
1970            tcg_out_insn(s, RX, C, TCG_REG_R0, TCG_TMP0, TCG_REG_NONE, ofs);
1971        } else {
1972            tcg_out_insn(s, RXY, CG, TCG_REG_R0, TCG_TMP0, TCG_REG_NONE, ofs);
1973        }
1974
1975        tcg_out16(s, RI_BRC | (S390_CC_NE << 4));
1976        ldst->label_ptr[0] = s->code_ptr++;
1977
1978        h->index = TCG_TMP0;
1979        tcg_out_insn(s, RXY, LG, h->index, TCG_TMP0, TCG_REG_NONE,
1980                     offsetof(CPUTLBEntry, addend));
1981
1982        if (addr_type == TCG_TYPE_I32) {
1983            tcg_out_insn(s, RRE, ALGFR, h->index, addr_reg);
1984            h->base = TCG_REG_NONE;
1985        } else {
1986            h->base = addr_reg;
1987        }
1988        h->disp = 0;
1989    } else {
1990        if (a_mask) {
1991            ldst = new_ldst_label(s);
1992            ldst->is_ld = is_ld;
1993            ldst->oi = oi;
1994            ldst->addr_reg = addr_reg;
1995
1996            tcg_debug_assert(a_mask <= 0xffff);
1997            tcg_out_insn(s, RI, TMLL, addr_reg, a_mask);
1998
1999            tcg_out16(s, RI_BRC | (S390_TM_NE << 4));
2000            ldst->label_ptr[0] = s->code_ptr++;
2001        }
2002
2003        h->base = addr_reg;
2004        if (addr_type == TCG_TYPE_I32) {
2005            tcg_out_ext32u(s, TCG_TMP0, addr_reg);
2006            h->base = TCG_TMP0;
2007        }
2008        if (guest_base < 0x80000) {
2009            h->index = TCG_REG_NONE;
2010            h->disp = guest_base;
2011        } else {
2012            h->index = TCG_GUEST_BASE_REG;
2013            h->disp = 0;
2014        }
2015    }
2016
2017    return ldst;
2018}
2019
2020static void tcg_out_qemu_ld(TCGContext* s, TCGReg data_reg, TCGReg addr_reg,
2021                            MemOpIdx oi, TCGType data_type)
2022{
2023    TCGLabelQemuLdst *ldst;
2024    HostAddress h;
2025
2026    ldst = prepare_host_addr(s, &h, addr_reg, oi, true);
2027    tcg_out_qemu_ld_direct(s, get_memop(oi), data_reg, h);
2028
2029    if (ldst) {
2030        ldst->type = data_type;
2031        ldst->datalo_reg = data_reg;
2032        ldst->raddr = tcg_splitwx_to_rx(s->code_ptr);
2033    }
2034}
2035
2036static void tcg_out_qemu_st(TCGContext* s, TCGReg data_reg, TCGReg addr_reg,
2037                            MemOpIdx oi, TCGType data_type)
2038{
2039    TCGLabelQemuLdst *ldst;
2040    HostAddress h;
2041
2042    ldst = prepare_host_addr(s, &h, addr_reg, oi, false);
2043    tcg_out_qemu_st_direct(s, get_memop(oi), data_reg, h);
2044
2045    if (ldst) {
2046        ldst->type = data_type;
2047        ldst->datalo_reg = data_reg;
2048        ldst->raddr = tcg_splitwx_to_rx(s->code_ptr);
2049    }
2050}
2051
2052static void tcg_out_qemu_ldst_i128(TCGContext *s, TCGReg datalo, TCGReg datahi,
2053                                   TCGReg addr_reg, MemOpIdx oi, bool is_ld)
2054{
2055    TCGLabel *l1 = NULL, *l2 = NULL;
2056    TCGLabelQemuLdst *ldst;
2057    HostAddress h;
2058    bool need_bswap;
2059    bool use_pair;
2060    S390Opcode insn;
2061
2062    ldst = prepare_host_addr(s, &h, addr_reg, oi, is_ld);
2063
2064    use_pair = h.aa.atom < MO_128;
2065    need_bswap = get_memop(oi) & MO_BSWAP;
2066
2067    if (!use_pair) {
2068        /*
2069         * Atomicity requires we use LPQ.  If we've already checked for
2070         * 16-byte alignment, that's all we need.  If we arrive with
2071         * lesser alignment, we have determined that less than 16-byte
2072         * alignment can be satisfied with two 8-byte loads.
2073         */
2074        if (h.aa.align < MO_128) {
2075            use_pair = true;
2076            l1 = gen_new_label();
2077            l2 = gen_new_label();
2078
2079            tcg_out_insn(s, RI, TMLL, addr_reg, 15);
2080            tgen_branch(s, S390_TM_NE, l1);
2081        }
2082
2083        tcg_debug_assert(!need_bswap);
2084        tcg_debug_assert(datalo & 1);
2085        tcg_debug_assert(datahi == datalo - 1);
2086        insn = is_ld ? RXY_LPQ : RXY_STPQ;
2087        tcg_out_insn_RXY(s, insn, datahi, h.base, h.index, h.disp);
2088
2089        if (use_pair) {
2090            tgen_branch(s, S390_CC_ALWAYS, l2);
2091            tcg_out_label(s, l1);
2092        }
2093    }
2094    if (use_pair) {
2095        TCGReg d1, d2;
2096
2097        if (need_bswap) {
2098            d1 = datalo, d2 = datahi;
2099            insn = is_ld ? RXY_LRVG : RXY_STRVG;
2100        } else {
2101            d1 = datahi, d2 = datalo;
2102            insn = is_ld ? RXY_LG : RXY_STG;
2103        }
2104
2105        if (h.base == d1 || h.index == d1) {
2106            tcg_out_insn(s, RXY, LAY, TCG_TMP0, h.base, h.index, h.disp);
2107            h.base = TCG_TMP0;
2108            h.index = TCG_REG_NONE;
2109            h.disp = 0;
2110        }
2111        tcg_out_insn_RXY(s, insn, d1, h.base, h.index, h.disp);
2112        tcg_out_insn_RXY(s, insn, d2, h.base, h.index, h.disp + 8);
2113    }
2114    if (l2) {
2115        tcg_out_label(s, l2);
2116    }
2117
2118    if (ldst) {
2119        ldst->type = TCG_TYPE_I128;
2120        ldst->datalo_reg = datalo;
2121        ldst->datahi_reg = datahi;
2122        ldst->raddr = tcg_splitwx_to_rx(s->code_ptr);
2123    }
2124}
2125
2126static void tcg_out_exit_tb(TCGContext *s, uintptr_t a0)
2127{
2128    /* Reuse the zeroing that exists for goto_ptr.  */
2129    if (a0 == 0) {
2130        tgen_gotoi(s, S390_CC_ALWAYS, tcg_code_gen_epilogue);
2131    } else {
2132        tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R2, a0);
2133        tgen_gotoi(s, S390_CC_ALWAYS, tb_ret_addr);
2134    }
2135}
2136
2137static void tcg_out_goto_tb(TCGContext *s, int which)
2138{
2139    /*
2140     * Branch displacement must be aligned for atomic patching;
2141     * see if we need to add extra nop before branch
2142     */
2143    if (!QEMU_PTR_IS_ALIGNED(s->code_ptr + 1, 4)) {
2144        tcg_out16(s, NOP);
2145    }
2146    tcg_out16(s, RIL_BRCL | (S390_CC_ALWAYS << 4));
2147    set_jmp_insn_offset(s, which);
2148    s->code_ptr += 2;
2149    set_jmp_reset_offset(s, which);
2150}
2151
2152void tb_target_set_jmp_target(const TranslationBlock *tb, int n,
2153                              uintptr_t jmp_rx, uintptr_t jmp_rw)
2154{
2155    if (!HAVE_FACILITY(GEN_INST_EXT)) {
2156        return;
2157    }
2158    /* patch the branch destination */
2159    uintptr_t addr = tb->jmp_target_addr[n];
2160    intptr_t disp = addr - (jmp_rx - 2);
2161    qatomic_set((int32_t *)jmp_rw, disp / 2);
2162    /* no need to flush icache explicitly */
2163}
2164
2165
2166static void tgen_add(TCGContext *s, TCGType type,
2167                     TCGReg a0, TCGReg a1, TCGReg a2)
2168{
2169    if (a0 != a1) {
2170        tcg_out_insn(s, RX, LA, a0, a1, a2, 0);
2171    } else if (type == TCG_TYPE_I32) {
2172        tcg_out_insn(s, RR, AR, a0, a2);
2173    } else {
2174        tcg_out_insn(s, RRE, AGR, a0, a2);
2175    }
2176}
2177
2178static void tgen_addi(TCGContext *s, TCGType type,
2179                      TCGReg a0, TCGReg a1, tcg_target_long a2)
2180{
2181    if (a0 == a1) {
2182        if (type == TCG_TYPE_I32) {
2183            if (a2 == (int16_t)a2) {
2184                tcg_out_insn(s, RI, AHI, a0, a2);
2185            } else {
2186                tcg_out_insn(s, RIL, AFI, a0, a2);
2187            }
2188            return;
2189        }
2190        if (a2 == (int16_t)a2) {
2191            tcg_out_insn(s, RI, AGHI, a0, a2);
2192            return;
2193        }
2194        if (a2 == (int32_t)a2) {
2195            tcg_out_insn(s, RIL, AGFI, a0, a2);
2196            return;
2197        }
2198        if (a2 == (uint32_t)a2) {
2199            tcg_out_insn(s, RIL, ALGFI, a0, a2);
2200            return;
2201        }
2202        if (-a2 == (uint32_t)-a2) {
2203            tcg_out_insn(s, RIL, SLGFI, a0, -a2);
2204            return;
2205        }
2206    }
2207    tcg_out_mem(s, RX_LA, RXY_LAY, a0, a1, TCG_REG_NONE, a2);
2208}
2209
2210static const TCGOutOpBinary outop_add = {
2211    .base.static_constraint = C_O1_I2(r, r, ri),
2212    .out_rrr = tgen_add,
2213    .out_rri = tgen_addi,
2214};
2215
2216static void tgen_and(TCGContext *s, TCGType type,
2217                     TCGReg a0, TCGReg a1, TCGReg a2)
2218{
2219    if (type != TCG_TYPE_I32) {
2220        tcg_out_insn(s, RRFa, NGRK, a0, a1, a2);
2221    } else if (a0 == a1) {
2222        tcg_out_insn(s, RR, NR, a0, a2);
2223    } else {
2224        tcg_out_insn(s, RRFa, NRK, a0, a1, a2);
2225    }
2226}
2227
2228static void tgen_andi_3(TCGContext *s, TCGType type,
2229                        TCGReg a0, TCGReg a1, tcg_target_long a2)
2230{
2231    tcg_out_mov(s, type, a0, a1);
2232    tgen_andi(s, type, a0, a2);
2233}
2234
2235static const TCGOutOpBinary outop_and = {
2236    .base.static_constraint = C_O1_I2(r, r, rNKR),
2237    .out_rrr = tgen_and,
2238    .out_rri = tgen_andi_3,
2239};
2240
2241static void tgen_andc(TCGContext *s, TCGType type,
2242                      TCGReg a0, TCGReg a1, TCGReg a2)
2243{
2244    if (type == TCG_TYPE_I32) {
2245        tcg_out_insn(s, RRFa, NCRK, a0, a1, a2);
2246    } else {
2247        tcg_out_insn(s, RRFa, NCGRK, a0, a1, a2);
2248    }
2249}
2250
2251static TCGConstraintSetIndex cset_misc3_rrr(TCGType type, unsigned flags)
2252{
2253    return HAVE_FACILITY(MISC_INSN_EXT3) ? C_O1_I2(r, r, r) : C_NotImplemented;
2254}
2255
2256static const TCGOutOpBinary outop_andc = {
2257    .base.static_constraint = C_Dynamic,
2258    .base.dynamic_constraint = cset_misc3_rrr,
2259    .out_rrr = tgen_andc,
2260};
2261
2262static void tgen_clz_int(TCGContext *s, TCGReg dest, TCGReg a1,
2263                         TCGArg a2, int a2const)
2264{
2265    /*
2266     * Since this sets both R and R+1, we have no choice but to store the
2267     * result into R0, allowing R1 == TCG_TMP0 to be clobbered as well.
2268     */
2269    QEMU_BUILD_BUG_ON(TCG_TMP0 != TCG_REG_R1);
2270    tcg_out_insn(s, RRE, FLOGR, TCG_REG_R0, a1);
2271
2272    if (a2const && a2 == 64) {
2273        tcg_out_mov(s, TCG_TYPE_I64, dest, TCG_REG_R0);
2274        return;
2275    }
2276
2277    /*
2278     * Conditions from FLOGR are:
2279     *   2 -> one bit found
2280     *   8 -> no one bit found
2281     */
2282    tgen_movcond_int(s, TCG_TYPE_I64, dest, a2, a2const, TCG_REG_R0, 8, 2);
2283}
2284
2285static void tgen_clz(TCGContext *s, TCGType type,
2286                     TCGReg a0, TCGReg a1, TCGReg a2)
2287{
2288    tgen_clz_int(s, a0, a1, a2, false);
2289}
2290
2291static void tgen_clzi(TCGContext *s, TCGType type,
2292                      TCGReg a0, TCGReg a1, tcg_target_long a2)
2293{
2294    tgen_clz_int(s, a0, a1, a2, true);
2295}
2296
2297static TCGConstraintSetIndex cset_clz(TCGType type, unsigned flags)
2298{
2299    return type == TCG_TYPE_I64 ? C_O1_I2(r, r, rI) : C_NotImplemented;
2300}
2301
2302static const TCGOutOpBinary outop_clz = {
2303    .base.static_constraint = C_Dynamic,
2304    .base.dynamic_constraint = cset_clz,
2305    .out_rrr = tgen_clz,
2306    .out_rri = tgen_clzi,
2307};
2308
2309static void tgen_ctpop(TCGContext *s, TCGType type, TCGReg dest, TCGReg src)
2310{
2311    /* With MIE3, and bit 0 of m4 set, we get the complete result. */
2312    if (HAVE_FACILITY(MISC_INSN_EXT3)) {
2313        if (type == TCG_TYPE_I32) {
2314            tcg_out_ext32u(s, dest, src);
2315            src = dest;
2316        }
2317        tcg_out_insn(s, RRFc, POPCNT, dest, src, 8);
2318        return;
2319    }
2320
2321    /* Without MIE3, each byte gets the count of bits for the byte. */
2322    tcg_out_insn(s, RRFc, POPCNT, dest, src, 0);
2323
2324    /* Multiply to sum each byte at the top of the word. */
2325    if (type == TCG_TYPE_I32) {
2326        tcg_out_insn(s, RIL, MSFI, dest, 0x01010101);
2327        tcg_out_sh32(s, RS_SRL, dest, TCG_REG_NONE, 24);
2328    } else {
2329        tcg_out_movi(s, TCG_TYPE_I64, TCG_TMP0, 0x0101010101010101ull);
2330        tcg_out_insn(s, RRE, MSGR, dest, TCG_TMP0);
2331        tcg_out_sh64(s, RSY_SRLG, dest, dest, TCG_REG_NONE, 56);
2332    }
2333}
2334
2335static const TCGOutOpUnary outop_ctpop = {
2336    .base.static_constraint = C_O1_I1(r, r),
2337    .out_rr = tgen_ctpop,
2338};
2339
2340static const TCGOutOpBinary outop_ctz = {
2341    .base.static_constraint = C_NotImplemented,
2342};
2343
2344static const TCGOutOpBinary outop_divs = {
2345    .base.static_constraint = C_NotImplemented,
2346};
2347
2348static void tgen_divs2(TCGContext *s, TCGType type,
2349                       TCGReg a0, TCGReg a1, TCGReg a4)
2350{
2351    tcg_debug_assert((a1 & 1) == 0);
2352    tcg_debug_assert(a0 == a1 + 1);
2353    if (type == TCG_TYPE_I32) {
2354        tcg_out_insn(s, RR, DR, a1, a4);
2355    } else {
2356        /*
2357         * TODO: Move the sign-extend of the numerator from a2 into a3
2358         * into the tcg backend, instead of in early expansion.  It is
2359         * required for 32-bit DR, but not 64-bit DSGR.
2360         */
2361        tcg_out_insn(s, RRE, DSGR, a1, a4);
2362    }
2363}
2364
2365static const TCGOutOpDivRem outop_divs2 = {
2366    .base.static_constraint = C_O2_I3(o, m, 0, 1, r),
2367    .out_rr01r = tgen_divs2,
2368};
2369
2370static const TCGOutOpBinary outop_divu = {
2371    .base.static_constraint = C_NotImplemented,
2372};
2373
2374static void tgen_divu2(TCGContext *s, TCGType type,
2375                       TCGReg a0, TCGReg a1, TCGReg a4)
2376{
2377    tcg_debug_assert((a1 & 1) == 0);
2378    tcg_debug_assert(a0 == a1 + 1);
2379    if (type == TCG_TYPE_I32) {
2380        tcg_out_insn(s, RRE, DLR, a1, a4);
2381    } else {
2382        tcg_out_insn(s, RRE, DLGR, a1, a4);
2383    }
2384}
2385
2386static const TCGOutOpDivRem outop_divu2 = {
2387    .base.static_constraint = C_O2_I3(o, m, 0, 1, r),
2388    .out_rr01r = tgen_divu2,
2389};
2390
2391static void tgen_eqv(TCGContext *s, TCGType type,
2392                      TCGReg a0, TCGReg a1, TCGReg a2)
2393{
2394    if (type == TCG_TYPE_I32) {
2395        tcg_out_insn(s, RRFa, NXRK, a0, a1, a2);
2396    } else {
2397        tcg_out_insn(s, RRFa, NXGRK, a0, a1, a2);
2398    }
2399}
2400
2401static const TCGOutOpBinary outop_eqv = {
2402    .base.static_constraint = C_Dynamic,
2403    .base.dynamic_constraint = cset_misc3_rrr,
2404    .out_rrr = tgen_eqv,
2405};
2406
2407static void tgen_mul(TCGContext *s, TCGType type,
2408                     TCGReg a0, TCGReg a1, TCGReg a2)
2409{
2410    if (type == TCG_TYPE_I32) {
2411        if (a0 == a1) {
2412            tcg_out_insn(s, RRE, MSR, a0, a2);
2413        } else {
2414            tcg_out_insn(s, RRFa, MSRKC, a0, a1, a2);
2415        }
2416    } else {
2417        if (a0 == a1) {
2418            tcg_out_insn(s, RRE, MSGR, a0, a2);
2419        } else {
2420            tcg_out_insn(s, RRFa, MSGRKC, a0, a1, a2);
2421        }
2422    }
2423}
2424
2425static void tgen_muli(TCGContext *s, TCGType type,
2426                      TCGReg a0, TCGReg a1, tcg_target_long a2)
2427{
2428    tcg_out_mov(s, type, a0, a1);
2429    if (type == TCG_TYPE_I32) {
2430        if (a2 == (int16_t)a2) {
2431            tcg_out_insn(s, RI, MHI, a0, a2);
2432        } else {
2433            tcg_out_insn(s, RIL, MSFI, a0, a2);
2434        }
2435    } else {
2436        if (a2 == (int16_t)a2) {
2437            tcg_out_insn(s, RI, MGHI, a0, a2);
2438        } else {
2439            tcg_out_insn(s, RIL, MSGFI, a0, a2);
2440        }
2441    }
2442}
2443
2444static TCGConstraintSetIndex cset_mul(TCGType type, unsigned flags)
2445{
2446    return (HAVE_FACILITY(MISC_INSN_EXT2)
2447            ? C_O1_I2(r, r, rJ)
2448            : C_O1_I2(r, 0, rJ));
2449}
2450
2451static const TCGOutOpBinary outop_mul = {
2452    .base.static_constraint = C_Dynamic,
2453    .base.dynamic_constraint = cset_mul,
2454    .out_rrr = tgen_mul,
2455    .out_rri = tgen_muli,
2456};
2457
2458static void tgen_muls2(TCGContext *s, TCGType type,
2459                       TCGReg a0, TCGReg a1, TCGReg a2, TCGReg a3)
2460{
2461    tcg_debug_assert((a1 & 1) == 0);
2462    tcg_debug_assert(a0 == a1 + 1);
2463    tcg_out_insn(s, RRFa, MGRK, a1, a2, a3);
2464}
2465
2466static TCGConstraintSetIndex cset_muls2(TCGType type, unsigned flags)
2467{
2468    return (type == TCG_TYPE_I64 && HAVE_FACILITY(MISC_INSN_EXT2)
2469            ? C_O2_I2(o, m, r, r) : C_NotImplemented);
2470}
2471
2472static const TCGOutOpMul2 outop_muls2 = {
2473    .base.static_constraint = C_Dynamic,
2474    .base.dynamic_constraint = cset_muls2,
2475    .out_rrrr = tgen_muls2,
2476};
2477
2478static const TCGOutOpBinary outop_mulsh = {
2479    .base.static_constraint = C_NotImplemented,
2480};
2481
2482static void tgen_mulu2(TCGContext *s, TCGType type,
2483                       TCGReg a0, TCGReg a1, TCGReg a2, TCGReg a3)
2484{
2485    tcg_debug_assert(a0 == a2);
2486    tcg_debug_assert((a1 & 1) == 0);
2487    tcg_debug_assert(a0 == a1 + 1);
2488    tcg_out_insn(s, RRE, MLGR, a1, a3);
2489}
2490
2491static TCGConstraintSetIndex cset_mulu2(TCGType type, unsigned flags)
2492{
2493    return (type == TCG_TYPE_I64 && HAVE_FACILITY(MISC_INSN_EXT2)
2494            ? C_O2_I2(o, m, 0, r) : C_NotImplemented);
2495}
2496
2497static const TCGOutOpMul2 outop_mulu2 = {
2498    .base.static_constraint = C_Dynamic,
2499    .base.dynamic_constraint = cset_mulu2,
2500    .out_rrrr = tgen_mulu2,
2501};
2502
2503static const TCGOutOpBinary outop_muluh = {
2504    .base.static_constraint = C_NotImplemented,
2505};
2506
2507static void tgen_nand(TCGContext *s, TCGType type,
2508                      TCGReg a0, TCGReg a1, TCGReg a2)
2509{
2510    if (type == TCG_TYPE_I32) {
2511        tcg_out_insn(s, RRFa, NNRK, a0, a1, a2);
2512    } else {
2513        tcg_out_insn(s, RRFa, NNGRK, a0, a1, a2);
2514    }
2515}
2516
2517static const TCGOutOpBinary outop_nand = {
2518    .base.static_constraint = C_Dynamic,
2519    .base.dynamic_constraint = cset_misc3_rrr,
2520    .out_rrr = tgen_nand,
2521};
2522
2523static void tgen_nor(TCGContext *s, TCGType type,
2524                      TCGReg a0, TCGReg a1, TCGReg a2)
2525{
2526    if (type == TCG_TYPE_I32) {
2527        tcg_out_insn(s, RRFa, NORK, a0, a1, a2);
2528    } else {
2529        tcg_out_insn(s, RRFa, NOGRK, a0, a1, a2);
2530    }
2531}
2532
2533static const TCGOutOpBinary outop_nor = {
2534    .base.static_constraint = C_Dynamic,
2535    .base.dynamic_constraint = cset_misc3_rrr,
2536    .out_rrr = tgen_nor,
2537};
2538
2539static void tgen_or(TCGContext *s, TCGType type,
2540                     TCGReg a0, TCGReg a1, TCGReg a2)
2541{
2542    if (type != TCG_TYPE_I32) {
2543        tcg_out_insn(s, RRFa, OGRK, a0, a1, a2);
2544    } else if (a0 == a1) {
2545        tcg_out_insn(s, RR, OR, a0, a2);
2546    } else {
2547        tcg_out_insn(s, RRFa, ORK, a0, a1, a2);
2548    }
2549}
2550
2551static void tgen_ori_3(TCGContext *s, TCGType type,
2552                        TCGReg a0, TCGReg a1, tcg_target_long a2)
2553{
2554    tcg_out_mov(s, type, a0, a1);
2555    tgen_ori(s, a0, type == TCG_TYPE_I32 ? (uint32_t)a2 : a2);
2556}
2557
2558static const TCGOutOpBinary outop_or = {
2559    .base.static_constraint = C_O1_I2(r, r, rK),
2560    .out_rrr = tgen_or,
2561    .out_rri = tgen_ori_3,
2562};
2563
2564static void tgen_orc(TCGContext *s, TCGType type,
2565                     TCGReg a0, TCGReg a1, TCGReg a2)
2566{
2567    if (type == TCG_TYPE_I32) {
2568        tcg_out_insn(s, RRFa, OCRK, a0, a1, a2);
2569    } else {
2570        tcg_out_insn(s, RRFa, OCGRK, a0, a1, a2);
2571    }
2572}
2573
2574static const TCGOutOpBinary outop_orc = {
2575    .base.static_constraint = C_Dynamic,
2576    .base.dynamic_constraint = cset_misc3_rrr,
2577    .out_rrr = tgen_orc,
2578};
2579
2580static const TCGOutOpBinary outop_rems = {
2581    .base.static_constraint = C_NotImplemented,
2582};
2583
2584static const TCGOutOpBinary outop_remu = {
2585    .base.static_constraint = C_NotImplemented,
2586};
2587
2588static void tgen_rotl_int(TCGContext *s, TCGType type, TCGReg dst,
2589                          TCGReg src, TCGReg v, tcg_target_long i)
2590{
2591    S390Opcode insn = type == TCG_TYPE_I32 ? RSY_RLL : RSY_RLLG;
2592    tcg_out_sh64(s, insn, dst, src, v, i);
2593}
2594
2595static void tgen_rotl(TCGContext *s, TCGType type,
2596                      TCGReg a0, TCGReg a1, TCGReg a2)
2597{
2598    tgen_rotl_int(s, type, a0, a1, a2, 0);
2599}
2600
2601static void tgen_rotli(TCGContext *s, TCGType type,
2602                       TCGReg a0, TCGReg a1, tcg_target_long a2)
2603{
2604    tgen_rotl_int(s, type, a0, a1, TCG_REG_NONE, a2);
2605}
2606
2607static const TCGOutOpBinary outop_rotl = {
2608    .base.static_constraint = C_O1_I2(r, r, ri),
2609    .out_rrr = tgen_rotl,
2610    .out_rri = tgen_rotli,
2611};
2612
2613static const TCGOutOpBinary outop_rotr = {
2614    .base.static_constraint = C_NotImplemented,
2615};
2616
2617static void tgen_sar_int(TCGContext *s, TCGType type, TCGReg dst,
2618                         TCGReg src, TCGReg v, tcg_target_long i)
2619{
2620    if (type != TCG_TYPE_I32) {
2621        tcg_out_sh64(s, RSY_SRAG, dst, src, v, i);
2622    } else if (dst == src) {
2623        tcg_out_sh32(s, RS_SRA, dst, v, i);
2624    } else {
2625        tcg_out_sh64(s, RSY_SRAK, dst, src, v, i);
2626    }
2627}
2628
2629static void tgen_sar(TCGContext *s, TCGType type,
2630                     TCGReg a0, TCGReg a1, TCGReg a2)
2631{
2632    tgen_sar_int(s, type, a0, a1, a2, 0);
2633}
2634
2635static void tgen_sari(TCGContext *s, TCGType type,
2636                      TCGReg a0, TCGReg a1, tcg_target_long a2)
2637{
2638    tgen_sar_int(s, type, a0, a1, TCG_REG_NONE, a2);
2639}
2640
2641static const TCGOutOpBinary outop_sar = {
2642    .base.static_constraint = C_O1_I2(r, r, ri),
2643    .out_rrr = tgen_sar,
2644    .out_rri = tgen_sari,
2645};
2646
2647static void tgen_shl_int(TCGContext *s, TCGType type, TCGReg dst,
2648                         TCGReg src, TCGReg v, tcg_target_long i)
2649{
2650    if (type != TCG_TYPE_I32) {
2651        tcg_out_sh64(s, RSY_SLLG, dst, src, v, i);
2652    } else if (dst == src) {
2653        tcg_out_sh32(s, RS_SLL, dst, v, i);
2654    } else {
2655        tcg_out_sh64(s, RSY_SLLK, dst, src, v, i);
2656    }
2657}
2658
2659static void tgen_shl(TCGContext *s, TCGType type,
2660                     TCGReg a0, TCGReg a1, TCGReg a2)
2661{
2662    tgen_shl_int(s, type, a0, a1, a2, 0);
2663}
2664
2665static void tgen_shli(TCGContext *s, TCGType type,
2666                      TCGReg a0, TCGReg a1, tcg_target_long a2)
2667{
2668    tgen_shl_int(s, type, a0, a1, TCG_REG_NONE, a2);
2669}
2670
2671static const TCGOutOpBinary outop_shl = {
2672    .base.static_constraint = C_O1_I2(r, r, ri),
2673    .out_rrr = tgen_shl,
2674    .out_rri = tgen_shli,
2675};
2676
2677static void tgen_shr_int(TCGContext *s, TCGType type, TCGReg dst,
2678                         TCGReg src, TCGReg v, tcg_target_long i)
2679{
2680    if (type != TCG_TYPE_I32) {
2681        tcg_out_sh64(s, RSY_SRLG, dst, src, v, i);
2682    } else if (dst == src) {
2683        tcg_out_sh32(s, RS_SRL, dst, v, i);
2684    } else {
2685        tcg_out_sh64(s, RSY_SRLK, dst, src, v, i);
2686    }
2687}
2688
2689static void tgen_shr(TCGContext *s, TCGType type,
2690                     TCGReg a0, TCGReg a1, TCGReg a2)
2691{
2692    tgen_shr_int(s, type, a0, a1, a2, 0);
2693}
2694
2695static void tgen_shri(TCGContext *s, TCGType type,
2696                      TCGReg a0, TCGReg a1, tcg_target_long a2)
2697{
2698    tgen_shr_int(s, type, a0, a1, TCG_REG_NONE, a2);
2699}
2700
2701static const TCGOutOpBinary outop_shr = {
2702    .base.static_constraint = C_O1_I2(r, r, ri),
2703    .out_rrr = tgen_shr,
2704    .out_rri = tgen_shri,
2705};
2706
2707static void tgen_sub(TCGContext *s, TCGType type,
2708                     TCGReg a0, TCGReg a1, TCGReg a2)
2709{
2710    if (type != TCG_TYPE_I32) {
2711        tcg_out_insn(s, RRFa, SGRK, a0, a1, a2);
2712    } else if (a0 == a1) {
2713        tcg_out_insn(s, RR, SR, a0, a2);
2714    } else {
2715        tcg_out_insn(s, RRFa, SRK, a0, a1, a2);
2716    }
2717}
2718
2719static const TCGOutOpSubtract outop_sub = {
2720    .base.static_constraint = C_O1_I2(r, r, r),
2721    .out_rrr = tgen_sub,
2722};
2723
2724static void tgen_xor(TCGContext *s, TCGType type,
2725                     TCGReg a0, TCGReg a1, TCGReg a2)
2726{
2727    if (type != TCG_TYPE_I32) {
2728        tcg_out_insn(s, RRFa, XGRK, a0, a1, a2);
2729    } else if (a0 == a1) {
2730        tcg_out_insn(s, RR, XR, a0, a2);
2731    } else {
2732        tcg_out_insn(s, RRFa, XRK, a0, a1, a2);
2733    }
2734}
2735
2736static void tgen_xori_3(TCGContext *s, TCGType type,
2737                        TCGReg a0, TCGReg a1, tcg_target_long a2)
2738{
2739    tcg_out_mov(s, type, a0, a1);
2740    tgen_xori(s, a0, type == TCG_TYPE_I32 ? (uint32_t)a2 : a2);
2741}
2742
2743static const TCGOutOpBinary outop_xor = {
2744    .base.static_constraint = C_O1_I2(r, r, rK),
2745    .out_rrr = tgen_xor,
2746    .out_rri = tgen_xori_3,
2747};
2748
2749static void tgen_bswap16(TCGContext *s, TCGType type,
2750                         TCGReg a0, TCGReg a1, unsigned flags)
2751{
2752    if (type == TCG_TYPE_I32) {
2753        tcg_out_insn(s, RRE, LRVR, a0, a1);
2754        tcg_out_sh32(s, (flags & TCG_BSWAP_OS ? RS_SRA : RS_SRL),
2755                     a0, TCG_REG_NONE, 16);
2756    } else {
2757        tcg_out_insn(s, RRE, LRVGR, a0, a1);
2758        tcg_out_sh64(s, (flags & TCG_BSWAP_OS ? RSY_SRAG : RSY_SRLG),
2759                     a0, a0, TCG_REG_NONE, 48);
2760    }
2761}
2762
2763static const TCGOutOpBswap outop_bswap16 = {
2764    .base.static_constraint = C_O1_I1(r, r),
2765    .out_rr = tgen_bswap16,
2766};
2767
2768static void tgen_bswap32(TCGContext *s, TCGType type,
2769                         TCGReg a0, TCGReg a1, unsigned flags)
2770{
2771    tcg_out_insn(s, RRE, LRVR, a0, a1);
2772    if (flags & TCG_BSWAP_OS) {
2773        tcg_out_ext32s(s, a0, a0);
2774    } else if ((flags & (TCG_BSWAP_IZ | TCG_BSWAP_OZ)) == TCG_BSWAP_OZ) {
2775        tcg_out_ext32u(s, a0, a0);
2776    }
2777}
2778
2779static const TCGOutOpBswap outop_bswap32 = {
2780    .base.static_constraint = C_O1_I1(r, r),
2781    .out_rr = tgen_bswap32,
2782};
2783
2784static void tgen_bswap64(TCGContext *s, TCGType type, TCGReg a0, TCGReg a1)
2785{
2786    tcg_out_insn(s, RRE, LRVGR, a0, a1);
2787}
2788
2789static const TCGOutOpUnary outop_bswap64 = {
2790    .base.static_constraint = C_O1_I1(r, r),
2791    .out_rr = tgen_bswap64,
2792};
2793
2794static void tgen_neg(TCGContext *s, TCGType type, TCGReg a0, TCGReg a1)
2795{
2796    if (type == TCG_TYPE_I32) {
2797        tcg_out_insn(s, RR, LCR, a0, a1);
2798    } else {
2799        tcg_out_insn(s, RRE, LCGR, a0, a1);
2800    }
2801}
2802
2803static const TCGOutOpUnary outop_neg = {
2804    .base.static_constraint = C_O1_I1(r, r),
2805    .out_rr = tgen_neg,
2806};
2807
2808static void tgen_not(TCGContext *s, TCGType type, TCGReg a0, TCGReg a1)
2809{
2810    tgen_nor(s, type, a0, a1, a1);
2811}
2812
2813static TCGConstraintSetIndex cset_not(TCGType type, unsigned flags)
2814{
2815    return HAVE_FACILITY(MISC_INSN_EXT3) ? C_O1_I1(r, r) : C_NotImplemented;
2816}
2817
2818static const TCGOutOpUnary outop_not = {
2819    .base.static_constraint = C_Dynamic,
2820    .base.dynamic_constraint = cset_not,
2821    .out_rr = tgen_not,
2822};
2823
2824
2825# define OP_32_64(x) \
2826        case glue(glue(INDEX_op_,x),_i32): \
2827        case glue(glue(INDEX_op_,x),_i64)
2828
2829static void tcg_out_op(TCGContext *s, TCGOpcode opc, TCGType type,
2830                       const TCGArg args[TCG_MAX_OP_ARGS],
2831                       const int const_args[TCG_MAX_OP_ARGS])
2832{
2833    TCGArg a0, a1, a2;
2834
2835    switch (opc) {
2836    case INDEX_op_goto_ptr:
2837        a0 = args[0];
2838        tcg_out_insn(s, RR, BCR, S390_CC_ALWAYS, a0);
2839        break;
2840
2841    OP_32_64(ld8u):
2842        /* ??? LLC (RXY format) is only present with the extended-immediate
2843           facility, whereas LLGC is always present.  */
2844        tcg_out_mem(s, 0, RXY_LLGC, args[0], args[1], TCG_REG_NONE, args[2]);
2845        break;
2846
2847    OP_32_64(ld8s):
2848        /* ??? LB is no smaller than LGB, so no point to using it.  */
2849        tcg_out_mem(s, 0, RXY_LGB, args[0], args[1], TCG_REG_NONE, args[2]);
2850        break;
2851
2852    OP_32_64(ld16u):
2853        /* ??? LLH (RXY format) is only present with the extended-immediate
2854           facility, whereas LLGH is always present.  */
2855        tcg_out_mem(s, 0, RXY_LLGH, args[0], args[1], TCG_REG_NONE, args[2]);
2856        break;
2857
2858    case INDEX_op_ld16s_i32:
2859        tcg_out_mem(s, RX_LH, RXY_LHY, args[0], args[1], TCG_REG_NONE, args[2]);
2860        break;
2861
2862    case INDEX_op_ld_i32:
2863        tcg_out_ld(s, TCG_TYPE_I32, args[0], args[1], args[2]);
2864        break;
2865
2866    OP_32_64(st8):
2867        tcg_out_mem(s, RX_STC, RXY_STCY, args[0], args[1],
2868                    TCG_REG_NONE, args[2]);
2869        break;
2870
2871    OP_32_64(st16):
2872        tcg_out_mem(s, RX_STH, RXY_STHY, args[0], args[1],
2873                    TCG_REG_NONE, args[2]);
2874        break;
2875
2876    case INDEX_op_st_i32:
2877        tcg_out_st(s, TCG_TYPE_I32, args[0], args[1], args[2]);
2878        break;
2879
2880    case INDEX_op_add2_i32:
2881        if (const_args[4]) {
2882            tcg_out_insn(s, RIL, ALFI, args[0], args[4]);
2883        } else {
2884            tcg_out_insn(s, RR, ALR, args[0], args[4]);
2885        }
2886        tcg_out_insn(s, RRE, ALCR, args[1], args[5]);
2887        break;
2888    case INDEX_op_sub2_i32:
2889        if (const_args[4]) {
2890            tcg_out_insn(s, RIL, SLFI, args[0], args[4]);
2891        } else {
2892            tcg_out_insn(s, RR, SLR, args[0], args[4]);
2893        }
2894        tcg_out_insn(s, RRE, SLBR, args[1], args[5]);
2895        break;
2896
2897    case INDEX_op_br:
2898        tgen_branch(s, S390_CC_ALWAYS, arg_label(args[0]));
2899        break;
2900
2901    case INDEX_op_qemu_ld_i32:
2902        tcg_out_qemu_ld(s, args[0], args[1], args[2], TCG_TYPE_I32);
2903        break;
2904    case INDEX_op_qemu_ld_i64:
2905        tcg_out_qemu_ld(s, args[0], args[1], args[2], TCG_TYPE_I64);
2906        break;
2907    case INDEX_op_qemu_st_i32:
2908        tcg_out_qemu_st(s, args[0], args[1], args[2], TCG_TYPE_I32);
2909        break;
2910    case INDEX_op_qemu_st_i64:
2911        tcg_out_qemu_st(s, args[0], args[1], args[2], TCG_TYPE_I64);
2912        break;
2913    case INDEX_op_qemu_ld_i128:
2914        tcg_out_qemu_ldst_i128(s, args[0], args[1], args[2], args[3], true);
2915        break;
2916    case INDEX_op_qemu_st_i128:
2917        tcg_out_qemu_ldst_i128(s, args[0], args[1], args[2], args[3], false);
2918        break;
2919
2920    case INDEX_op_ld16s_i64:
2921        tcg_out_mem(s, 0, RXY_LGH, args[0], args[1], TCG_REG_NONE, args[2]);
2922        break;
2923    case INDEX_op_ld32u_i64:
2924        tcg_out_mem(s, 0, RXY_LLGF, args[0], args[1], TCG_REG_NONE, args[2]);
2925        break;
2926    case INDEX_op_ld32s_i64:
2927        tcg_out_mem(s, 0, RXY_LGF, args[0], args[1], TCG_REG_NONE, args[2]);
2928        break;
2929    case INDEX_op_ld_i64:
2930        tcg_out_ld(s, TCG_TYPE_I64, args[0], args[1], args[2]);
2931        break;
2932
2933    case INDEX_op_st32_i64:
2934        tcg_out_st(s, TCG_TYPE_I32, args[0], args[1], args[2]);
2935        break;
2936    case INDEX_op_st_i64:
2937        tcg_out_st(s, TCG_TYPE_I64, args[0], args[1], args[2]);
2938        break;
2939
2940    case INDEX_op_add2_i64:
2941        if (const_args[4]) {
2942            if ((int64_t)args[4] >= 0) {
2943                tcg_out_insn(s, RIL, ALGFI, args[0], args[4]);
2944            } else {
2945                tcg_out_insn(s, RIL, SLGFI, args[0], -args[4]);
2946            }
2947        } else {
2948            tcg_out_insn(s, RRE, ALGR, args[0], args[4]);
2949        }
2950        tcg_out_insn(s, RRE, ALCGR, args[1], args[5]);
2951        break;
2952    case INDEX_op_sub2_i64:
2953        if (const_args[4]) {
2954            if ((int64_t)args[4] >= 0) {
2955                tcg_out_insn(s, RIL, SLGFI, args[0], args[4]);
2956            } else {
2957                tcg_out_insn(s, RIL, ALGFI, args[0], -args[4]);
2958            }
2959        } else {
2960            tcg_out_insn(s, RRE, SLGR, args[0], args[4]);
2961        }
2962        tcg_out_insn(s, RRE, SLBGR, args[1], args[5]);
2963        break;
2964
2965    OP_32_64(deposit):
2966        a0 = args[0], a1 = args[1], a2 = args[2];
2967        if (const_args[1]) {
2968            tgen_deposit(s, a0, a2, args[3], args[4], 1);
2969        } else {
2970            /* Since we can't support "0Z" as a constraint, we allow a1 in
2971               any register.  Fix things up as if a matching constraint.  */
2972            if (a0 != a1) {
2973                if (a0 == a2) {
2974                    tcg_out_mov(s, type, TCG_TMP0, a2);
2975                    a2 = TCG_TMP0;
2976                }
2977                tcg_out_mov(s, type, a0, a1);
2978            }
2979            tgen_deposit(s, a0, a2, args[3], args[4], 0);
2980        }
2981        break;
2982
2983    OP_32_64(sextract):
2984        tgen_sextract(s, args[0], args[1], args[2], args[3]);
2985        break;
2986
2987    case INDEX_op_mb:
2988        /* The host memory model is quite strong, we simply need to
2989           serialize the instruction stream.  */
2990        if (args[0] & TCG_MO_ST_LD) {
2991            /* fast-bcr-serialization facility (45) is present */
2992            tcg_out_insn(s, RR, BCR, 14, 0);
2993        }
2994        break;
2995
2996    case INDEX_op_call:     /* Always emitted via tcg_out_call.  */
2997    case INDEX_op_exit_tb:  /* Always emitted via tcg_out_exit_tb.  */
2998    case INDEX_op_goto_tb:  /* Always emitted via tcg_out_goto_tb.  */
2999    case INDEX_op_ext_i32_i64:  /* Always emitted via tcg_reg_alloc_op.  */
3000    case INDEX_op_extu_i32_i64:
3001    case INDEX_op_extrl_i64_i32:
3002    default:
3003        g_assert_not_reached();
3004    }
3005}
3006
3007static bool tcg_out_dup_vec(TCGContext *s, TCGType type, unsigned vece,
3008                            TCGReg dst, TCGReg src)
3009{
3010    if (is_general_reg(src)) {
3011        /* Replicate general register into two MO_64. */
3012        tcg_out_insn(s, VRRf, VLVGP, dst, src, src);
3013        if (vece == MO_64) {
3014            return true;
3015        }
3016        src = dst;
3017    }
3018
3019    /*
3020     * Recall that the "standard" integer, within a vector, is the
3021     * rightmost element of the leftmost doubleword, a-la VLLEZ.
3022     */
3023    tcg_out_insn(s, VRIc, VREP, dst, (8 >> vece) - 1, src, vece);
3024    return true;
3025}
3026
3027static bool tcg_out_dupm_vec(TCGContext *s, TCGType type, unsigned vece,
3028                             TCGReg dst, TCGReg base, intptr_t offset)
3029{
3030    tcg_out_vrx_mem(s, VRX_VLREP, dst, base, TCG_REG_NONE, offset, vece);
3031    return true;
3032}
3033
3034static void tcg_out_dupi_vec(TCGContext *s, TCGType type, unsigned vece,
3035                             TCGReg dst, int64_t val)
3036{
3037    int i, mask, msb, lsb;
3038
3039    /* Look for int16_t elements.  */
3040    if (vece <= MO_16 ||
3041        (vece == MO_32 ? (int32_t)val : val) == (int16_t)val) {
3042        tcg_out_insn(s, VRIa, VREPI, dst, val, vece);
3043        return;
3044    }
3045
3046    /* Look for bit masks.  */
3047    if (vece == MO_32) {
3048        if (risbg_mask((int32_t)val)) {
3049            /* Handle wraparound by swapping msb and lsb.  */
3050            if ((val & 0x80000001u) == 0x80000001u) {
3051                msb = 32 - ctz32(~val);
3052                lsb = clz32(~val) - 1;
3053            } else {
3054                msb = clz32(val);
3055                lsb = 31 - ctz32(val);
3056            }
3057            tcg_out_insn(s, VRIb, VGM, dst, msb, lsb, MO_32);
3058            return;
3059        }
3060    } else {
3061        if (risbg_mask(val)) {
3062            /* Handle wraparound by swapping msb and lsb.  */
3063            if ((val & 0x8000000000000001ull) == 0x8000000000000001ull) {
3064                /* Handle wraparound by swapping msb and lsb.  */
3065                msb = 64 - ctz64(~val);
3066                lsb = clz64(~val) - 1;
3067            } else {
3068                msb = clz64(val);
3069                lsb = 63 - ctz64(val);
3070            }
3071            tcg_out_insn(s, VRIb, VGM, dst, msb, lsb, MO_64);
3072            return;
3073        }
3074    }
3075
3076    /* Look for all bytes 0x00 or 0xff.  */
3077    for (i = mask = 0; i < 8; i++) {
3078        uint8_t byte = val >> (i * 8);
3079        if (byte == 0xff) {
3080            mask |= 1 << i;
3081        } else if (byte != 0) {
3082            break;
3083        }
3084    }
3085    if (i == 8) {
3086        tcg_out_insn(s, VRIa, VGBM, dst, mask * 0x0101, 0);
3087        return;
3088    }
3089
3090    /* Otherwise, stuff it in the constant pool.  */
3091    tcg_out_insn(s, RIL, LARL, TCG_TMP0, 0);
3092    new_pool_label(s, val, R_390_PC32DBL, s->code_ptr - 2, 2);
3093    tcg_out_insn(s, VRX, VLREP, dst, TCG_TMP0, TCG_REG_NONE, 0, MO_64);
3094}
3095
3096static bool tcg_out_cmp_vec_noinv(TCGContext *s, unsigned vece, TCGReg a0,
3097                                  TCGReg a1, TCGReg a2, TCGCond cond)
3098{
3099    bool need_swap = false, need_inv = false;
3100
3101    switch (cond) {
3102    case TCG_COND_EQ:
3103    case TCG_COND_GT:
3104    case TCG_COND_GTU:
3105        break;
3106    case TCG_COND_NE:
3107    case TCG_COND_LE:
3108    case TCG_COND_LEU:
3109        need_inv = true;
3110        break;
3111    case TCG_COND_LT:
3112    case TCG_COND_LTU:
3113        need_swap = true;
3114        break;
3115    case TCG_COND_GE:
3116    case TCG_COND_GEU:
3117        need_swap = need_inv = true;
3118        break;
3119    default:
3120        g_assert_not_reached();
3121    }
3122
3123    if (need_inv) {
3124        cond = tcg_invert_cond(cond);
3125    }
3126    if (need_swap) {
3127        TCGReg swap = a1;
3128        a1 = a2;
3129        a2 = swap;
3130        cond = tcg_swap_cond(cond);
3131    }
3132
3133    switch (cond) {
3134    case TCG_COND_EQ:
3135        tcg_out_insn(s, VRRc, VCEQ, a0, a1, a2, vece);
3136        break;
3137    case TCG_COND_GT:
3138        tcg_out_insn(s, VRRc, VCH, a0, a1, a2, vece);
3139        break;
3140    case TCG_COND_GTU:
3141        tcg_out_insn(s, VRRc, VCHL, a0, a1, a2, vece);
3142        break;
3143    default:
3144        g_assert_not_reached();
3145    }
3146    return need_inv;
3147}
3148
3149static void tcg_out_cmp_vec(TCGContext *s, unsigned vece, TCGReg a0,
3150                            TCGReg a1, TCGReg a2, TCGCond cond)
3151{
3152    if (tcg_out_cmp_vec_noinv(s, vece, a0, a1, a2, cond)) {
3153        tcg_out_insn(s, VRRc, VNO, a0, a0, a0, 0);
3154    }
3155}
3156
3157static void tcg_out_cmpsel_vec(TCGContext *s, unsigned vece, TCGReg a0,
3158                               TCGReg c1, TCGReg c2, TCGArg v3,
3159                               int const_v3, TCGReg v4, TCGCond cond)
3160{
3161    bool inv = tcg_out_cmp_vec_noinv(s, vece, TCG_VEC_TMP0, c1, c2, cond);
3162
3163    if (!const_v3) {
3164        if (inv) {
3165            tcg_out_insn(s, VRRe, VSEL, a0, v4, v3, TCG_VEC_TMP0);
3166        } else {
3167            tcg_out_insn(s, VRRe, VSEL, a0, v3, v4, TCG_VEC_TMP0);
3168        }
3169    } else if (v3) {
3170        if (inv) {
3171            tcg_out_insn(s, VRRc, VOC, a0, v4, TCG_VEC_TMP0, 0);
3172        } else {
3173            tcg_out_insn(s, VRRc, VO, a0, v4, TCG_VEC_TMP0, 0);
3174        }
3175    } else {
3176        if (inv) {
3177            tcg_out_insn(s, VRRc, VN, a0, v4, TCG_VEC_TMP0, 0);
3178        } else {
3179            tcg_out_insn(s, VRRc, VNC, a0, v4, TCG_VEC_TMP0, 0);
3180        }
3181    }
3182}
3183
3184static void tcg_out_vec_op(TCGContext *s, TCGOpcode opc,
3185                           unsigned vecl, unsigned vece,
3186                           const TCGArg args[TCG_MAX_OP_ARGS],
3187                           const int const_args[TCG_MAX_OP_ARGS])
3188{
3189    TCGType type = vecl + TCG_TYPE_V64;
3190    TCGArg a0 = args[0], a1 = args[1], a2 = args[2];
3191
3192    switch (opc) {
3193    case INDEX_op_ld_vec:
3194        tcg_out_ld(s, type, a0, a1, a2);
3195        break;
3196    case INDEX_op_st_vec:
3197        tcg_out_st(s, type, a0, a1, a2);
3198        break;
3199    case INDEX_op_dupm_vec:
3200        tcg_out_dupm_vec(s, type, vece, a0, a1, a2);
3201        break;
3202
3203    case INDEX_op_abs_vec:
3204        tcg_out_insn(s, VRRa, VLP, a0, a1, vece);
3205        break;
3206    case INDEX_op_neg_vec:
3207        tcg_out_insn(s, VRRa, VLC, a0, a1, vece);
3208        break;
3209    case INDEX_op_not_vec:
3210        tcg_out_insn(s, VRRc, VNO, a0, a1, a1, 0);
3211        break;
3212
3213    case INDEX_op_add_vec:
3214        tcg_out_insn(s, VRRc, VA, a0, a1, a2, vece);
3215        break;
3216    case INDEX_op_sub_vec:
3217        tcg_out_insn(s, VRRc, VS, a0, a1, a2, vece);
3218        break;
3219    case INDEX_op_and_vec:
3220        tcg_out_insn(s, VRRc, VN, a0, a1, a2, 0);
3221        break;
3222    case INDEX_op_andc_vec:
3223        tcg_out_insn(s, VRRc, VNC, a0, a1, a2, 0);
3224        break;
3225    case INDEX_op_mul_vec:
3226        tcg_out_insn(s, VRRc, VML, a0, a1, a2, vece);
3227        break;
3228    case INDEX_op_or_vec:
3229        tcg_out_insn(s, VRRc, VO, a0, a1, a2, 0);
3230        break;
3231    case INDEX_op_orc_vec:
3232        tcg_out_insn(s, VRRc, VOC, a0, a1, a2, 0);
3233        break;
3234    case INDEX_op_xor_vec:
3235        tcg_out_insn(s, VRRc, VX, a0, a1, a2, 0);
3236        break;
3237    case INDEX_op_nand_vec:
3238        tcg_out_insn(s, VRRc, VNN, a0, a1, a2, 0);
3239        break;
3240    case INDEX_op_nor_vec:
3241        tcg_out_insn(s, VRRc, VNO, a0, a1, a2, 0);
3242        break;
3243    case INDEX_op_eqv_vec:
3244        tcg_out_insn(s, VRRc, VNX, a0, a1, a2, 0);
3245        break;
3246
3247    case INDEX_op_shli_vec:
3248        tcg_out_insn(s, VRSa, VESL, a0, a2, TCG_REG_NONE, a1, vece);
3249        break;
3250    case INDEX_op_shri_vec:
3251        tcg_out_insn(s, VRSa, VESRL, a0, a2, TCG_REG_NONE, a1, vece);
3252        break;
3253    case INDEX_op_sari_vec:
3254        tcg_out_insn(s, VRSa, VESRA, a0, a2, TCG_REG_NONE, a1, vece);
3255        break;
3256    case INDEX_op_rotli_vec:
3257        tcg_out_insn(s, VRSa, VERLL, a0, a2, TCG_REG_NONE, a1, vece);
3258        break;
3259    case INDEX_op_shls_vec:
3260        tcg_out_insn(s, VRSa, VESL, a0, 0, a2, a1, vece);
3261        break;
3262    case INDEX_op_shrs_vec:
3263        tcg_out_insn(s, VRSa, VESRL, a0, 0, a2, a1, vece);
3264        break;
3265    case INDEX_op_sars_vec:
3266        tcg_out_insn(s, VRSa, VESRA, a0, 0, a2, a1, vece);
3267        break;
3268    case INDEX_op_rotls_vec:
3269        tcg_out_insn(s, VRSa, VERLL, a0, 0, a2, a1, vece);
3270        break;
3271    case INDEX_op_shlv_vec:
3272        tcg_out_insn(s, VRRc, VESLV, a0, a1, a2, vece);
3273        break;
3274    case INDEX_op_shrv_vec:
3275        tcg_out_insn(s, VRRc, VESRLV, a0, a1, a2, vece);
3276        break;
3277    case INDEX_op_sarv_vec:
3278        tcg_out_insn(s, VRRc, VESRAV, a0, a1, a2, vece);
3279        break;
3280    case INDEX_op_rotlv_vec:
3281        tcg_out_insn(s, VRRc, VERLLV, a0, a1, a2, vece);
3282        break;
3283
3284    case INDEX_op_smin_vec:
3285        tcg_out_insn(s, VRRc, VMN, a0, a1, a2, vece);
3286        break;
3287    case INDEX_op_smax_vec:
3288        tcg_out_insn(s, VRRc, VMX, a0, a1, a2, vece);
3289        break;
3290    case INDEX_op_umin_vec:
3291        tcg_out_insn(s, VRRc, VMNL, a0, a1, a2, vece);
3292        break;
3293    case INDEX_op_umax_vec:
3294        tcg_out_insn(s, VRRc, VMXL, a0, a1, a2, vece);
3295        break;
3296
3297    case INDEX_op_bitsel_vec:
3298        tcg_out_insn(s, VRRe, VSEL, a0, a2, args[3], a1);
3299        break;
3300
3301    case INDEX_op_cmp_vec:
3302        tcg_out_cmp_vec(s, vece, a0, a1, a2, args[3]);
3303        break;
3304    case INDEX_op_cmpsel_vec:
3305        tcg_out_cmpsel_vec(s, vece, a0, a1, a2, args[3], const_args[3],
3306                           args[4], args[5]);
3307        break;
3308
3309    case INDEX_op_s390_vuph_vec:
3310        tcg_out_insn(s, VRRa, VUPH, a0, a1, vece);
3311        break;
3312    case INDEX_op_s390_vupl_vec:
3313        tcg_out_insn(s, VRRa, VUPL, a0, a1, vece);
3314        break;
3315    case INDEX_op_s390_vpks_vec:
3316        tcg_out_insn(s, VRRc, VPKS, a0, a1, a2, vece);
3317        break;
3318
3319    case INDEX_op_mov_vec:   /* Always emitted via tcg_out_mov.  */
3320    case INDEX_op_dup_vec:   /* Always emitted via tcg_out_dup_vec.  */
3321    default:
3322        g_assert_not_reached();
3323    }
3324}
3325
3326int tcg_can_emit_vec_op(TCGOpcode opc, TCGType type, unsigned vece)
3327{
3328    switch (opc) {
3329    case INDEX_op_abs_vec:
3330    case INDEX_op_add_vec:
3331    case INDEX_op_and_vec:
3332    case INDEX_op_andc_vec:
3333    case INDEX_op_bitsel_vec:
3334    case INDEX_op_eqv_vec:
3335    case INDEX_op_nand_vec:
3336    case INDEX_op_neg_vec:
3337    case INDEX_op_nor_vec:
3338    case INDEX_op_not_vec:
3339    case INDEX_op_or_vec:
3340    case INDEX_op_orc_vec:
3341    case INDEX_op_rotli_vec:
3342    case INDEX_op_rotls_vec:
3343    case INDEX_op_rotlv_vec:
3344    case INDEX_op_sari_vec:
3345    case INDEX_op_sars_vec:
3346    case INDEX_op_sarv_vec:
3347    case INDEX_op_shli_vec:
3348    case INDEX_op_shls_vec:
3349    case INDEX_op_shlv_vec:
3350    case INDEX_op_shri_vec:
3351    case INDEX_op_shrs_vec:
3352    case INDEX_op_shrv_vec:
3353    case INDEX_op_smax_vec:
3354    case INDEX_op_smin_vec:
3355    case INDEX_op_sub_vec:
3356    case INDEX_op_umax_vec:
3357    case INDEX_op_umin_vec:
3358    case INDEX_op_xor_vec:
3359    case INDEX_op_cmp_vec:
3360    case INDEX_op_cmpsel_vec:
3361        return 1;
3362    case INDEX_op_rotrv_vec:
3363        return -1;
3364    case INDEX_op_mul_vec:
3365        return vece < MO_64;
3366    case INDEX_op_ssadd_vec:
3367    case INDEX_op_sssub_vec:
3368        return vece < MO_64 ? -1 : 0;
3369    default:
3370        return 0;
3371    }
3372}
3373
3374static void expand_vec_sat(TCGType type, unsigned vece, TCGv_vec v0,
3375                           TCGv_vec v1, TCGv_vec v2, TCGOpcode add_sub_opc)
3376{
3377    TCGv_vec h1 = tcg_temp_new_vec(type);
3378    TCGv_vec h2 = tcg_temp_new_vec(type);
3379    TCGv_vec l1 = tcg_temp_new_vec(type);
3380    TCGv_vec l2 = tcg_temp_new_vec(type);
3381
3382    tcg_debug_assert (vece < MO_64);
3383
3384    /* Unpack with sign-extension. */
3385    vec_gen_2(INDEX_op_s390_vuph_vec, type, vece,
3386              tcgv_vec_arg(h1), tcgv_vec_arg(v1));
3387    vec_gen_2(INDEX_op_s390_vuph_vec, type, vece,
3388              tcgv_vec_arg(h2), tcgv_vec_arg(v2));
3389
3390    vec_gen_2(INDEX_op_s390_vupl_vec, type, vece,
3391              tcgv_vec_arg(l1), tcgv_vec_arg(v1));
3392    vec_gen_2(INDEX_op_s390_vupl_vec, type, vece,
3393              tcgv_vec_arg(l2), tcgv_vec_arg(v2));
3394
3395    /* Arithmetic on a wider element size. */
3396    vec_gen_3(add_sub_opc, type, vece + 1, tcgv_vec_arg(h1),
3397              tcgv_vec_arg(h1), tcgv_vec_arg(h2));
3398    vec_gen_3(add_sub_opc, type, vece + 1, tcgv_vec_arg(l1),
3399              tcgv_vec_arg(l1), tcgv_vec_arg(l2));
3400
3401    /* Pack with saturation. */
3402    vec_gen_3(INDEX_op_s390_vpks_vec, type, vece + 1,
3403              tcgv_vec_arg(v0), tcgv_vec_arg(h1), tcgv_vec_arg(l1));
3404
3405    tcg_temp_free_vec(h1);
3406    tcg_temp_free_vec(h2);
3407    tcg_temp_free_vec(l1);
3408    tcg_temp_free_vec(l2);
3409}
3410
3411void tcg_expand_vec_op(TCGOpcode opc, TCGType type, unsigned vece,
3412                       TCGArg a0, ...)
3413{
3414    va_list va;
3415    TCGv_vec v0, v1, v2, t0;
3416
3417    va_start(va, a0);
3418    v0 = temp_tcgv_vec(arg_temp(a0));
3419    v1 = temp_tcgv_vec(arg_temp(va_arg(va, TCGArg)));
3420    v2 = temp_tcgv_vec(arg_temp(va_arg(va, TCGArg)));
3421
3422    switch (opc) {
3423    case INDEX_op_rotrv_vec:
3424        t0 = tcg_temp_new_vec(type);
3425        tcg_gen_neg_vec(vece, t0, v2);
3426        tcg_gen_rotlv_vec(vece, v0, v1, t0);
3427        tcg_temp_free_vec(t0);
3428        break;
3429
3430    case INDEX_op_ssadd_vec:
3431        expand_vec_sat(type, vece, v0, v1, v2, INDEX_op_add_vec);
3432        break;
3433    case INDEX_op_sssub_vec:
3434        expand_vec_sat(type, vece, v0, v1, v2, INDEX_op_sub_vec);
3435        break;
3436
3437    default:
3438        g_assert_not_reached();
3439    }
3440    va_end(va);
3441}
3442
3443static TCGConstraintSetIndex
3444tcg_target_op_def(TCGOpcode op, TCGType type, unsigned flags)
3445{
3446    switch (op) {
3447    case INDEX_op_goto_ptr:
3448        return C_O0_I1(r);
3449
3450    case INDEX_op_ld8u_i32:
3451    case INDEX_op_ld8u_i64:
3452    case INDEX_op_ld8s_i32:
3453    case INDEX_op_ld8s_i64:
3454    case INDEX_op_ld16u_i32:
3455    case INDEX_op_ld16u_i64:
3456    case INDEX_op_ld16s_i32:
3457    case INDEX_op_ld16s_i64:
3458    case INDEX_op_ld_i32:
3459    case INDEX_op_ld32u_i64:
3460    case INDEX_op_ld32s_i64:
3461    case INDEX_op_ld_i64:
3462        return C_O1_I1(r, r);
3463
3464    case INDEX_op_st8_i32:
3465    case INDEX_op_st8_i64:
3466    case INDEX_op_st16_i32:
3467    case INDEX_op_st16_i64:
3468    case INDEX_op_st_i32:
3469    case INDEX_op_st32_i64:
3470    case INDEX_op_st_i64:
3471        return C_O0_I2(r, r);
3472
3473    case INDEX_op_ext_i32_i64:
3474    case INDEX_op_extu_i32_i64:
3475    case INDEX_op_sextract_i32:
3476    case INDEX_op_sextract_i64:
3477        return C_O1_I1(r, r);
3478
3479    case INDEX_op_qemu_ld_i32:
3480    case INDEX_op_qemu_ld_i64:
3481        return C_O1_I1(r, r);
3482    case INDEX_op_qemu_st_i64:
3483    case INDEX_op_qemu_st_i32:
3484        return C_O0_I2(r, r);
3485    case INDEX_op_qemu_ld_i128:
3486        return C_O2_I1(o, m, r);
3487    case INDEX_op_qemu_st_i128:
3488        return C_O0_I3(o, m, r);
3489
3490    case INDEX_op_deposit_i32:
3491    case INDEX_op_deposit_i64:
3492        return C_O1_I2(r, rZ, r);
3493
3494    case INDEX_op_add2_i32:
3495    case INDEX_op_sub2_i32:
3496        return C_N1_O1_I4(r, r, 0, 1, ri, r);
3497
3498    case INDEX_op_add2_i64:
3499    case INDEX_op_sub2_i64:
3500        return C_N1_O1_I4(r, r, 0, 1, rJU, r);
3501
3502    case INDEX_op_st_vec:
3503        return C_O0_I2(v, r);
3504    case INDEX_op_ld_vec:
3505    case INDEX_op_dupm_vec:
3506        return C_O1_I1(v, r);
3507    case INDEX_op_dup_vec:
3508        return C_O1_I1(v, vr);
3509    case INDEX_op_abs_vec:
3510    case INDEX_op_neg_vec:
3511    case INDEX_op_not_vec:
3512    case INDEX_op_rotli_vec:
3513    case INDEX_op_sari_vec:
3514    case INDEX_op_shli_vec:
3515    case INDEX_op_shri_vec:
3516    case INDEX_op_s390_vuph_vec:
3517    case INDEX_op_s390_vupl_vec:
3518        return C_O1_I1(v, v);
3519    case INDEX_op_add_vec:
3520    case INDEX_op_sub_vec:
3521    case INDEX_op_and_vec:
3522    case INDEX_op_andc_vec:
3523    case INDEX_op_or_vec:
3524    case INDEX_op_orc_vec:
3525    case INDEX_op_xor_vec:
3526    case INDEX_op_nand_vec:
3527    case INDEX_op_nor_vec:
3528    case INDEX_op_eqv_vec:
3529    case INDEX_op_cmp_vec:
3530    case INDEX_op_mul_vec:
3531    case INDEX_op_rotlv_vec:
3532    case INDEX_op_rotrv_vec:
3533    case INDEX_op_shlv_vec:
3534    case INDEX_op_shrv_vec:
3535    case INDEX_op_sarv_vec:
3536    case INDEX_op_smax_vec:
3537    case INDEX_op_smin_vec:
3538    case INDEX_op_umax_vec:
3539    case INDEX_op_umin_vec:
3540    case INDEX_op_s390_vpks_vec:
3541        return C_O1_I2(v, v, v);
3542    case INDEX_op_rotls_vec:
3543    case INDEX_op_shls_vec:
3544    case INDEX_op_shrs_vec:
3545    case INDEX_op_sars_vec:
3546        return C_O1_I2(v, v, r);
3547    case INDEX_op_bitsel_vec:
3548        return C_O1_I3(v, v, v, v);
3549    case INDEX_op_cmpsel_vec:
3550        return (TCG_TARGET_HAS_orc_vec
3551                ? C_O1_I4(v, v, v, vZM, v)
3552                : C_O1_I4(v, v, v, vZ, v));
3553
3554    default:
3555        return C_NotImplemented;
3556    }
3557}
3558
3559/*
3560 * Mainline glibc added HWCAP_S390_VX before it was kernel abi.
3561 * Some distros have fixed this up locally, others have not.
3562 */
3563#ifndef HWCAP_S390_VXRS
3564#define HWCAP_S390_VXRS 2048
3565#endif
3566
3567static void query_s390_facilities(void)
3568{
3569    unsigned long hwcap = qemu_getauxval(AT_HWCAP);
3570    const char *which;
3571
3572    /* Is STORE FACILITY LIST EXTENDED available?  Honestly, I believe this
3573       is present on all 64-bit systems, but let's check for it anyway.  */
3574    if (hwcap & HWCAP_S390_STFLE) {
3575        register int r0 __asm__("0") = ARRAY_SIZE(s390_facilities) - 1;
3576        register void *r1 __asm__("1") = s390_facilities;
3577
3578        /* stfle 0(%r1) */
3579        asm volatile(".word 0xb2b0,0x1000"
3580                     : "=r"(r0) : "r"(r0), "r"(r1) : "memory", "cc");
3581    }
3582
3583    /*
3584     * Use of vector registers requires os support beyond the facility bit.
3585     * If the kernel does not advertise support, disable the facility bits.
3586     * There is nothing else we currently care about in the 3rd word, so
3587     * disable VECTOR with one store.
3588     */
3589    if (!(hwcap & HWCAP_S390_VXRS)) {
3590        s390_facilities[2] = 0;
3591    }
3592
3593    /*
3594     * Minimum supported cpu revision is z196.
3595     * Check for all required facilities.
3596     * ZARCH_ACTIVE is done via preprocessor check for 64-bit.
3597     */
3598    if (!HAVE_FACILITY(LONG_DISP)) {
3599        which = "long-displacement";
3600        goto fail;
3601    }
3602    if (!HAVE_FACILITY(EXT_IMM)) {
3603        which = "extended-immediate";
3604        goto fail;
3605    }
3606    if (!HAVE_FACILITY(GEN_INST_EXT)) {
3607        which = "general-instructions-extension";
3608        goto fail;
3609    }
3610    /*
3611     * Facility 45 is a big bin that contains: distinct-operands,
3612     * fast-BCR-serialization, high-word, population-count,
3613     * interlocked-access-1, and load/store-on-condition-1
3614     */
3615    if (!HAVE_FACILITY(45)) {
3616        which = "45";
3617        goto fail;
3618    }
3619    return;
3620
3621 fail:
3622    error_report("%s: missing required facility %s", __func__, which);
3623    exit(EXIT_FAILURE);
3624}
3625
3626static void tcg_target_init(TCGContext *s)
3627{
3628    query_s390_facilities();
3629
3630    tcg_target_available_regs[TCG_TYPE_I32] = 0xffff;
3631    tcg_target_available_regs[TCG_TYPE_I64] = 0xffff;
3632    if (HAVE_FACILITY(VECTOR)) {
3633        tcg_target_available_regs[TCG_TYPE_V64] = 0xffffffff00000000ull;
3634        tcg_target_available_regs[TCG_TYPE_V128] = 0xffffffff00000000ull;
3635    }
3636
3637    tcg_target_call_clobber_regs = 0;
3638    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R0);
3639    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R1);
3640    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R2);
3641    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R3);
3642    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R4);
3643    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R5);
3644    /* The r6 register is technically call-saved, but it's also a parameter
3645       register, so it can get killed by setup for the qemu_st helper.  */
3646    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R6);
3647    /* The return register can be considered call-clobbered.  */
3648    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R14);
3649
3650    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V0);
3651    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V1);
3652    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V2);
3653    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V3);
3654    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V4);
3655    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V5);
3656    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V6);
3657    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V7);
3658    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V16);
3659    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V17);
3660    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V18);
3661    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V19);
3662    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V20);
3663    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V21);
3664    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V22);
3665    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V23);
3666    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V24);
3667    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V25);
3668    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V26);
3669    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V27);
3670    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V28);
3671    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V29);
3672    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V30);
3673    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V31);
3674
3675    s->reserved_regs = 0;
3676    tcg_regset_set_reg(s->reserved_regs, TCG_TMP0);
3677    tcg_regset_set_reg(s->reserved_regs, TCG_VEC_TMP0);
3678    /* XXX many insns can't be used with R0, so we better avoid it for now */
3679    tcg_regset_set_reg(s->reserved_regs, TCG_REG_R0);
3680    tcg_regset_set_reg(s->reserved_regs, TCG_REG_CALL_STACK);
3681}
3682
3683#define FRAME_SIZE  ((int)(TCG_TARGET_CALL_STACK_OFFSET          \
3684                           + TCG_STATIC_CALL_ARGS_SIZE           \
3685                           + CPU_TEMP_BUF_NLONGS * sizeof(long)))
3686
3687static void tcg_target_qemu_prologue(TCGContext *s)
3688{
3689    /* stmg %r6,%r15,48(%r15) (save registers) */
3690    tcg_out_insn(s, RXY, STMG, TCG_REG_R6, TCG_REG_R15, TCG_REG_R15, 48);
3691
3692    /* aghi %r15,-frame_size */
3693    tcg_out_insn(s, RI, AGHI, TCG_REG_R15, -FRAME_SIZE);
3694
3695    tcg_set_frame(s, TCG_REG_CALL_STACK,
3696                  TCG_STATIC_CALL_ARGS_SIZE + TCG_TARGET_CALL_STACK_OFFSET,
3697                  CPU_TEMP_BUF_NLONGS * sizeof(long));
3698
3699    if (!tcg_use_softmmu && guest_base >= 0x80000) {
3700        tcg_out_movi(s, TCG_TYPE_PTR, TCG_GUEST_BASE_REG, guest_base);
3701        tcg_regset_set_reg(s->reserved_regs, TCG_GUEST_BASE_REG);
3702    }
3703
3704    tcg_out_mov(s, TCG_TYPE_PTR, TCG_AREG0, tcg_target_call_iarg_regs[0]);
3705
3706    /* br %r3 (go to TB) */
3707    tcg_out_insn(s, RR, BCR, S390_CC_ALWAYS, tcg_target_call_iarg_regs[1]);
3708
3709    /*
3710     * Return path for goto_ptr. Set return value to 0, a-la exit_tb,
3711     * and fall through to the rest of the epilogue.
3712     */
3713    tcg_code_gen_epilogue = tcg_splitwx_to_rx(s->code_ptr);
3714    tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R2, 0);
3715
3716    /* TB epilogue */
3717    tb_ret_addr = tcg_splitwx_to_rx(s->code_ptr);
3718
3719    /* lmg %r6,%r15,fs+48(%r15) (restore registers) */
3720    tcg_out_insn(s, RXY, LMG, TCG_REG_R6, TCG_REG_R15, TCG_REG_R15,
3721                 FRAME_SIZE + 48);
3722
3723    /* br %r14 (return) */
3724    tcg_out_insn(s, RR, BCR, S390_CC_ALWAYS, TCG_REG_R14);
3725}
3726
3727static void tcg_out_tb_start(TCGContext *s)
3728{
3729    /* nothing to do */
3730}
3731
3732static void tcg_out_nop_fill(tcg_insn_unit *p, int count)
3733{
3734    memset(p, 0x07, count * sizeof(tcg_insn_unit));
3735}
3736
3737typedef struct {
3738    DebugFrameHeader h;
3739    uint8_t fde_def_cfa[4];
3740    uint8_t fde_reg_ofs[18];
3741} DebugFrame;
3742
3743/* We're expecting a 2 byte uleb128 encoded value.  */
3744QEMU_BUILD_BUG_ON(FRAME_SIZE >= (1 << 14));
3745
3746#define ELF_HOST_MACHINE  EM_S390
3747
3748static const DebugFrame debug_frame = {
3749    .h.cie.len = sizeof(DebugFrameCIE)-4, /* length after .len member */
3750    .h.cie.id = -1,
3751    .h.cie.version = 1,
3752    .h.cie.code_align = 1,
3753    .h.cie.data_align = 8,                /* sleb128 8 */
3754    .h.cie.return_column = TCG_REG_R14,
3755
3756    /* Total FDE size does not include the "len" member.  */
3757    .h.fde.len = sizeof(DebugFrame) - offsetof(DebugFrame, h.fde.cie_offset),
3758
3759    .fde_def_cfa = {
3760        12, TCG_REG_CALL_STACK,         /* DW_CFA_def_cfa %r15, ... */
3761        (FRAME_SIZE & 0x7f) | 0x80,     /* ... uleb128 FRAME_SIZE */
3762        (FRAME_SIZE >> 7)
3763    },
3764    .fde_reg_ofs = {
3765        0x86, 6,                        /* DW_CFA_offset, %r6, 48 */
3766        0x87, 7,                        /* DW_CFA_offset, %r7, 56 */
3767        0x88, 8,                        /* DW_CFA_offset, %r8, 64 */
3768        0x89, 9,                        /* DW_CFA_offset, %r92, 72 */
3769        0x8a, 10,                       /* DW_CFA_offset, %r10, 80 */
3770        0x8b, 11,                       /* DW_CFA_offset, %r11, 88 */
3771        0x8c, 12,                       /* DW_CFA_offset, %r12, 96 */
3772        0x8d, 13,                       /* DW_CFA_offset, %r13, 104 */
3773        0x8e, 14,                       /* DW_CFA_offset, %r14, 112 */
3774    }
3775};
3776
3777void tcg_register_jit(const void *buf, size_t buf_size)
3778{
3779    tcg_register_jit_int(buf, buf_size, &debug_frame, sizeof(debug_frame));
3780}
3781