xref: /openbmc/qemu/tcg/s390x/tcg-target.c.inc (revision 3ad5d4ccb4bdebdff4e90957bb2b8a93e5e418e2)
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, TCGReg dest, TCGReg src,
1567                         int ofs, int 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 void tgen_sextract(TCGContext *s, TCGReg dest, TCGReg src,
1586                          int ofs, int len)
1587{
1588    if (ofs == 0) {
1589        switch (len) {
1590        case 8:
1591            tcg_out_ext8s(s, TCG_TYPE_REG, dest, src);
1592            return;
1593        case 16:
1594            tcg_out_ext16s(s, TCG_TYPE_REG, dest, src);
1595            return;
1596        case 32:
1597            tcg_out_ext32s(s, dest, src);
1598            return;
1599        }
1600    }
1601    g_assert_not_reached();
1602}
1603
1604static void tgen_gotoi(TCGContext *s, int cc, const tcg_insn_unit *dest)
1605{
1606    ptrdiff_t off = tcg_pcrel_diff(s, dest) >> 1;
1607    if (off == (int16_t)off) {
1608        tcg_out_insn(s, RI, BRC, cc, off);
1609    } else if (off == (int32_t)off) {
1610        tcg_out_insn(s, RIL, BRCL, cc, off);
1611    } else {
1612        tcg_out_movi(s, TCG_TYPE_PTR, TCG_TMP0, (uintptr_t)dest);
1613        tcg_out_insn(s, RR, BCR, cc, TCG_TMP0);
1614    }
1615}
1616
1617static void tgen_branch(TCGContext *s, int cc, TCGLabel *l)
1618{
1619    if (l->has_value) {
1620        tgen_gotoi(s, cc, l->u.value_ptr);
1621    } else {
1622        tcg_out16(s, RI_BRC | (cc << 4));
1623        tcg_out_reloc(s, s->code_ptr, R_390_PC16DBL, l, 2);
1624        s->code_ptr += 1;
1625    }
1626}
1627
1628static void tgen_compare_branch(TCGContext *s, S390Opcode opc, int cc,
1629                                TCGReg r1, TCGReg r2, TCGLabel *l)
1630{
1631    tcg_out_reloc(s, s->code_ptr + 1, R_390_PC16DBL, l, 2);
1632    /* Format RIE-b */
1633    tcg_out16(s, (opc & 0xff00) | (r1 << 4) | r2);
1634    tcg_out16(s, 0);
1635    tcg_out16(s, cc << 12 | (opc & 0xff));
1636}
1637
1638static void tgen_compare_imm_branch(TCGContext *s, S390Opcode opc, int cc,
1639                                    TCGReg r1, int i2, TCGLabel *l)
1640{
1641    tcg_out_reloc(s, s->code_ptr + 1, R_390_PC16DBL, l, 2);
1642    /* Format RIE-c */
1643    tcg_out16(s, (opc & 0xff00) | (r1 << 4) | cc);
1644    tcg_out16(s, 0);
1645    tcg_out16(s, (i2 << 8) | (opc & 0xff));
1646}
1647
1648static void tgen_brcond(TCGContext *s, TCGType type, TCGCond c,
1649                        TCGReg r1, TCGArg c2, int c2const, TCGLabel *l)
1650{
1651    int cc;
1652
1653    if (!is_tst_cond(c)) {
1654        bool is_unsigned = is_unsigned_cond(c);
1655        bool in_range;
1656        S390Opcode opc;
1657
1658        cc = tcg_cond_to_s390_cond[c];
1659
1660        if (!c2const) {
1661            opc = (type == TCG_TYPE_I32
1662                   ? (is_unsigned ? RIEb_CLRJ : RIEb_CRJ)
1663                   : (is_unsigned ? RIEb_CLGRJ : RIEb_CGRJ));
1664            tgen_compare_branch(s, opc, cc, r1, c2, l);
1665            return;
1666        }
1667
1668        /*
1669         * COMPARE IMMEDIATE AND BRANCH RELATIVE has an 8-bit immediate field.
1670         * If the immediate we've been given does not fit that range, we'll
1671         * fall back to separate compare and branch instructions using the
1672         * larger comparison range afforded by COMPARE IMMEDIATE.
1673         */
1674        if (type == TCG_TYPE_I32) {
1675            if (is_unsigned) {
1676                opc = RIEc_CLIJ;
1677                in_range = (uint32_t)c2 == (uint8_t)c2;
1678            } else {
1679                opc = RIEc_CIJ;
1680                in_range = (int32_t)c2 == (int8_t)c2;
1681            }
1682        } else {
1683            if (is_unsigned) {
1684                opc = RIEc_CLGIJ;
1685                in_range = (uint64_t)c2 == (uint8_t)c2;
1686            } else {
1687                opc = RIEc_CGIJ;
1688                in_range = (int64_t)c2 == (int8_t)c2;
1689            }
1690        }
1691        if (in_range) {
1692            tgen_compare_imm_branch(s, opc, cc, r1, c2, l);
1693            return;
1694        }
1695    }
1696
1697    cc = tgen_cmp(s, type, c, r1, c2, c2const, false);
1698    tgen_branch(s, cc, l);
1699}
1700
1701static void tgen_brcondr(TCGContext *s, TCGType type, TCGCond c,
1702                         TCGReg a0, TCGReg a1, TCGLabel *l)
1703{
1704    tgen_brcond(s, type, c, a0, a1, false, l);
1705}
1706
1707static void tgen_brcondi(TCGContext *s, TCGType type, TCGCond c,
1708                         TCGReg a0, tcg_target_long a1, TCGLabel *l)
1709{
1710    tgen_brcond(s, type, c, a0, a1, true, l);
1711}
1712
1713static const TCGOutOpBrcond outop_brcond = {
1714    .base.static_constraint = C_O0_I2(r, rC),
1715    .out_rr = tgen_brcondr,
1716    .out_ri = tgen_brcondi,
1717};
1718
1719static void tcg_out_call_int(TCGContext *s, const tcg_insn_unit *dest)
1720{
1721    ptrdiff_t off = tcg_pcrel_diff(s, dest) >> 1;
1722    if (off == (int32_t)off) {
1723        tcg_out_insn(s, RIL, BRASL, TCG_REG_R14, off);
1724    } else {
1725        tcg_out_movi(s, TCG_TYPE_PTR, TCG_TMP0, (uintptr_t)dest);
1726        tcg_out_insn(s, RR, BASR, TCG_REG_R14, TCG_TMP0);
1727    }
1728}
1729
1730static void tcg_out_call(TCGContext *s, const tcg_insn_unit *dest,
1731                         const TCGHelperInfo *info)
1732{
1733    tcg_out_call_int(s, dest);
1734}
1735
1736typedef struct {
1737    TCGReg base;
1738    TCGReg index;
1739    int disp;
1740    TCGAtomAlign aa;
1741} HostAddress;
1742
1743bool tcg_target_has_memory_bswap(MemOp memop)
1744{
1745    TCGAtomAlign aa;
1746
1747    if ((memop & MO_SIZE) <= MO_64) {
1748        return true;
1749    }
1750
1751    /*
1752     * Reject 16-byte memop with 16-byte atomicity,
1753     * but do allow a pair of 64-bit operations.
1754     */
1755    aa = atom_and_align_for_opc(tcg_ctx, memop, MO_ATOM_IFALIGN, true);
1756    return aa.atom <= MO_64;
1757}
1758
1759static void tcg_out_qemu_ld_direct(TCGContext *s, MemOp opc, TCGReg data,
1760                                   HostAddress h)
1761{
1762    switch (opc & (MO_SSIZE | MO_BSWAP)) {
1763    case MO_UB:
1764        tcg_out_insn(s, RXY, LLGC, data, h.base, h.index, h.disp);
1765        break;
1766    case MO_SB:
1767        tcg_out_insn(s, RXY, LGB, data, h.base, h.index, h.disp);
1768        break;
1769
1770    case MO_UW | MO_BSWAP:
1771        /* swapped unsigned halfword load with upper bits zeroed */
1772        tcg_out_insn(s, RXY, LRVH, data, h.base, h.index, h.disp);
1773        tcg_out_ext16u(s, data, data);
1774        break;
1775    case MO_UW:
1776        tcg_out_insn(s, RXY, LLGH, data, h.base, h.index, h.disp);
1777        break;
1778
1779    case MO_SW | MO_BSWAP:
1780        /* swapped sign-extended halfword load */
1781        tcg_out_insn(s, RXY, LRVH, data, h.base, h.index, h.disp);
1782        tcg_out_ext16s(s, TCG_TYPE_REG, data, data);
1783        break;
1784    case MO_SW:
1785        tcg_out_insn(s, RXY, LGH, data, h.base, h.index, h.disp);
1786        break;
1787
1788    case MO_UL | MO_BSWAP:
1789        /* swapped unsigned int load with upper bits zeroed */
1790        tcg_out_insn(s, RXY, LRV, data, h.base, h.index, h.disp);
1791        tcg_out_ext32u(s, data, data);
1792        break;
1793    case MO_UL:
1794        tcg_out_insn(s, RXY, LLGF, data, h.base, h.index, h.disp);
1795        break;
1796
1797    case MO_SL | MO_BSWAP:
1798        /* swapped sign-extended int load */
1799        tcg_out_insn(s, RXY, LRV, data, h.base, h.index, h.disp);
1800        tcg_out_ext32s(s, data, data);
1801        break;
1802    case MO_SL:
1803        tcg_out_insn(s, RXY, LGF, data, h.base, h.index, h.disp);
1804        break;
1805
1806    case MO_UQ | MO_BSWAP:
1807        tcg_out_insn(s, RXY, LRVG, data, h.base, h.index, h.disp);
1808        break;
1809    case MO_UQ:
1810        tcg_out_insn(s, RXY, LG, data, h.base, h.index, h.disp);
1811        break;
1812
1813    default:
1814        g_assert_not_reached();
1815    }
1816}
1817
1818static void tcg_out_qemu_st_direct(TCGContext *s, MemOp opc, TCGReg data,
1819                                   HostAddress h)
1820{
1821    switch (opc & (MO_SIZE | MO_BSWAP)) {
1822    case MO_UB:
1823        if (h.disp >= 0 && h.disp < 0x1000) {
1824            tcg_out_insn(s, RX, STC, data, h.base, h.index, h.disp);
1825        } else {
1826            tcg_out_insn(s, RXY, STCY, data, h.base, h.index, h.disp);
1827        }
1828        break;
1829
1830    case MO_UW | MO_BSWAP:
1831        tcg_out_insn(s, RXY, STRVH, data, h.base, h.index, h.disp);
1832        break;
1833    case MO_UW:
1834        if (h.disp >= 0 && h.disp < 0x1000) {
1835            tcg_out_insn(s, RX, STH, data, h.base, h.index, h.disp);
1836        } else {
1837            tcg_out_insn(s, RXY, STHY, data, h.base, h.index, h.disp);
1838        }
1839        break;
1840
1841    case MO_UL | MO_BSWAP:
1842        tcg_out_insn(s, RXY, STRV, data, h.base, h.index, h.disp);
1843        break;
1844    case MO_UL:
1845        if (h.disp >= 0 && h.disp < 0x1000) {
1846            tcg_out_insn(s, RX, ST, data, h.base, h.index, h.disp);
1847        } else {
1848            tcg_out_insn(s, RXY, STY, data, h.base, h.index, h.disp);
1849        }
1850        break;
1851
1852    case MO_UQ | MO_BSWAP:
1853        tcg_out_insn(s, RXY, STRVG, data, h.base, h.index, h.disp);
1854        break;
1855    case MO_UQ:
1856        tcg_out_insn(s, RXY, STG, data, h.base, h.index, h.disp);
1857        break;
1858
1859    default:
1860        g_assert_not_reached();
1861    }
1862}
1863
1864static const TCGLdstHelperParam ldst_helper_param = {
1865    .ntmp = 1, .tmp = { TCG_TMP0 }
1866};
1867
1868static bool tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
1869{
1870    MemOp opc = get_memop(lb->oi);
1871
1872    if (!patch_reloc(lb->label_ptr[0], R_390_PC16DBL,
1873                     (intptr_t)tcg_splitwx_to_rx(s->code_ptr), 2)) {
1874        return false;
1875    }
1876
1877    tcg_out_ld_helper_args(s, lb, &ldst_helper_param);
1878    tcg_out_call_int(s, qemu_ld_helpers[opc & MO_SIZE]);
1879    tcg_out_ld_helper_ret(s, lb, false, &ldst_helper_param);
1880
1881    tgen_gotoi(s, S390_CC_ALWAYS, lb->raddr);
1882    return true;
1883}
1884
1885static bool tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
1886{
1887    MemOp opc = get_memop(lb->oi);
1888
1889    if (!patch_reloc(lb->label_ptr[0], R_390_PC16DBL,
1890                     (intptr_t)tcg_splitwx_to_rx(s->code_ptr), 2)) {
1891        return false;
1892    }
1893
1894    tcg_out_st_helper_args(s, lb, &ldst_helper_param);
1895    tcg_out_call_int(s, qemu_st_helpers[opc & MO_SIZE]);
1896
1897    tgen_gotoi(s, S390_CC_ALWAYS, lb->raddr);
1898    return true;
1899}
1900
1901/* We're expecting to use a 20-bit negative offset on the tlb memory ops.  */
1902#define MIN_TLB_MASK_TABLE_OFS  -(1 << 19)
1903
1904/*
1905 * For system-mode, perform the TLB load and compare.
1906 * For user-mode, perform any required alignment tests.
1907 * In both cases, return a TCGLabelQemuLdst structure if the slow path
1908 * is required and fill in @h with the host address for the fast path.
1909 */
1910static TCGLabelQemuLdst *prepare_host_addr(TCGContext *s, HostAddress *h,
1911                                           TCGReg addr_reg, MemOpIdx oi,
1912                                           bool is_ld)
1913{
1914    TCGType addr_type = s->addr_type;
1915    TCGLabelQemuLdst *ldst = NULL;
1916    MemOp opc = get_memop(oi);
1917    MemOp s_bits = opc & MO_SIZE;
1918    unsigned a_mask;
1919
1920    h->aa = atom_and_align_for_opc(s, opc, MO_ATOM_IFALIGN, s_bits == MO_128);
1921    a_mask = (1 << h->aa.align) - 1;
1922
1923    if (tcg_use_softmmu) {
1924        unsigned s_mask = (1 << s_bits) - 1;
1925        int mem_index = get_mmuidx(oi);
1926        int fast_off = tlb_mask_table_ofs(s, mem_index);
1927        int mask_off = fast_off + offsetof(CPUTLBDescFast, mask);
1928        int table_off = fast_off + offsetof(CPUTLBDescFast, table);
1929        int ofs, a_off;
1930        uint64_t tlb_mask;
1931
1932        ldst = new_ldst_label(s);
1933        ldst->is_ld = is_ld;
1934        ldst->oi = oi;
1935        ldst->addr_reg = addr_reg;
1936
1937        tcg_out_sh64(s, RSY_SRLG, TCG_TMP0, addr_reg, TCG_REG_NONE,
1938                     s->page_bits - CPU_TLB_ENTRY_BITS);
1939
1940        tcg_out_insn(s, RXY, NG, TCG_TMP0, TCG_AREG0, TCG_REG_NONE, mask_off);
1941        tcg_out_insn(s, RXY, AG, TCG_TMP0, TCG_AREG0, TCG_REG_NONE, table_off);
1942
1943        /*
1944         * For aligned accesses, we check the first byte and include the
1945         * alignment bits within the address.  For unaligned access, we
1946         * check that we don't cross pages using the address of the last
1947         * byte of the access.
1948         */
1949        a_off = (a_mask >= s_mask ? 0 : s_mask - a_mask);
1950        tlb_mask = (uint64_t)s->page_mask | a_mask;
1951        if (a_off == 0) {
1952            tgen_andi_risbg(s, TCG_REG_R0, addr_reg, tlb_mask);
1953        } else {
1954            tcg_out_insn(s, RX, LA, TCG_REG_R0, addr_reg, TCG_REG_NONE, a_off);
1955            tgen_andi(s, addr_type, TCG_REG_R0, tlb_mask);
1956        }
1957
1958        if (is_ld) {
1959            ofs = offsetof(CPUTLBEntry, addr_read);
1960        } else {
1961            ofs = offsetof(CPUTLBEntry, addr_write);
1962        }
1963        if (addr_type == TCG_TYPE_I32) {
1964            ofs += HOST_BIG_ENDIAN * 4;
1965            tcg_out_insn(s, RX, C, TCG_REG_R0, TCG_TMP0, TCG_REG_NONE, ofs);
1966        } else {
1967            tcg_out_insn(s, RXY, CG, TCG_REG_R0, TCG_TMP0, TCG_REG_NONE, ofs);
1968        }
1969
1970        tcg_out16(s, RI_BRC | (S390_CC_NE << 4));
1971        ldst->label_ptr[0] = s->code_ptr++;
1972
1973        h->index = TCG_TMP0;
1974        tcg_out_insn(s, RXY, LG, h->index, TCG_TMP0, TCG_REG_NONE,
1975                     offsetof(CPUTLBEntry, addend));
1976
1977        if (addr_type == TCG_TYPE_I32) {
1978            tcg_out_insn(s, RRE, ALGFR, h->index, addr_reg);
1979            h->base = TCG_REG_NONE;
1980        } else {
1981            h->base = addr_reg;
1982        }
1983        h->disp = 0;
1984    } else {
1985        if (a_mask) {
1986            ldst = new_ldst_label(s);
1987            ldst->is_ld = is_ld;
1988            ldst->oi = oi;
1989            ldst->addr_reg = addr_reg;
1990
1991            tcg_debug_assert(a_mask <= 0xffff);
1992            tcg_out_insn(s, RI, TMLL, addr_reg, a_mask);
1993
1994            tcg_out16(s, RI_BRC | (S390_TM_NE << 4));
1995            ldst->label_ptr[0] = s->code_ptr++;
1996        }
1997
1998        h->base = addr_reg;
1999        if (addr_type == TCG_TYPE_I32) {
2000            tcg_out_ext32u(s, TCG_TMP0, addr_reg);
2001            h->base = TCG_TMP0;
2002        }
2003        if (guest_base < 0x80000) {
2004            h->index = TCG_REG_NONE;
2005            h->disp = guest_base;
2006        } else {
2007            h->index = TCG_GUEST_BASE_REG;
2008            h->disp = 0;
2009        }
2010    }
2011
2012    return ldst;
2013}
2014
2015static void tcg_out_qemu_ld(TCGContext* s, TCGReg data_reg, TCGReg addr_reg,
2016                            MemOpIdx oi, TCGType data_type)
2017{
2018    TCGLabelQemuLdst *ldst;
2019    HostAddress h;
2020
2021    ldst = prepare_host_addr(s, &h, addr_reg, oi, true);
2022    tcg_out_qemu_ld_direct(s, get_memop(oi), data_reg, h);
2023
2024    if (ldst) {
2025        ldst->type = data_type;
2026        ldst->datalo_reg = data_reg;
2027        ldst->raddr = tcg_splitwx_to_rx(s->code_ptr);
2028    }
2029}
2030
2031static void tcg_out_qemu_st(TCGContext* s, TCGReg data_reg, TCGReg addr_reg,
2032                            MemOpIdx oi, TCGType data_type)
2033{
2034    TCGLabelQemuLdst *ldst;
2035    HostAddress h;
2036
2037    ldst = prepare_host_addr(s, &h, addr_reg, oi, false);
2038    tcg_out_qemu_st_direct(s, get_memop(oi), data_reg, h);
2039
2040    if (ldst) {
2041        ldst->type = data_type;
2042        ldst->datalo_reg = data_reg;
2043        ldst->raddr = tcg_splitwx_to_rx(s->code_ptr);
2044    }
2045}
2046
2047static void tcg_out_qemu_ldst_i128(TCGContext *s, TCGReg datalo, TCGReg datahi,
2048                                   TCGReg addr_reg, MemOpIdx oi, bool is_ld)
2049{
2050    TCGLabel *l1 = NULL, *l2 = NULL;
2051    TCGLabelQemuLdst *ldst;
2052    HostAddress h;
2053    bool need_bswap;
2054    bool use_pair;
2055    S390Opcode insn;
2056
2057    ldst = prepare_host_addr(s, &h, addr_reg, oi, is_ld);
2058
2059    use_pair = h.aa.atom < MO_128;
2060    need_bswap = get_memop(oi) & MO_BSWAP;
2061
2062    if (!use_pair) {
2063        /*
2064         * Atomicity requires we use LPQ.  If we've already checked for
2065         * 16-byte alignment, that's all we need.  If we arrive with
2066         * lesser alignment, we have determined that less than 16-byte
2067         * alignment can be satisfied with two 8-byte loads.
2068         */
2069        if (h.aa.align < MO_128) {
2070            use_pair = true;
2071            l1 = gen_new_label();
2072            l2 = gen_new_label();
2073
2074            tcg_out_insn(s, RI, TMLL, addr_reg, 15);
2075            tgen_branch(s, S390_TM_NE, l1);
2076        }
2077
2078        tcg_debug_assert(!need_bswap);
2079        tcg_debug_assert(datalo & 1);
2080        tcg_debug_assert(datahi == datalo - 1);
2081        insn = is_ld ? RXY_LPQ : RXY_STPQ;
2082        tcg_out_insn_RXY(s, insn, datahi, h.base, h.index, h.disp);
2083
2084        if (use_pair) {
2085            tgen_branch(s, S390_CC_ALWAYS, l2);
2086            tcg_out_label(s, l1);
2087        }
2088    }
2089    if (use_pair) {
2090        TCGReg d1, d2;
2091
2092        if (need_bswap) {
2093            d1 = datalo, d2 = datahi;
2094            insn = is_ld ? RXY_LRVG : RXY_STRVG;
2095        } else {
2096            d1 = datahi, d2 = datalo;
2097            insn = is_ld ? RXY_LG : RXY_STG;
2098        }
2099
2100        if (h.base == d1 || h.index == d1) {
2101            tcg_out_insn(s, RXY, LAY, TCG_TMP0, h.base, h.index, h.disp);
2102            h.base = TCG_TMP0;
2103            h.index = TCG_REG_NONE;
2104            h.disp = 0;
2105        }
2106        tcg_out_insn_RXY(s, insn, d1, h.base, h.index, h.disp);
2107        tcg_out_insn_RXY(s, insn, d2, h.base, h.index, h.disp + 8);
2108    }
2109    if (l2) {
2110        tcg_out_label(s, l2);
2111    }
2112
2113    if (ldst) {
2114        ldst->type = TCG_TYPE_I128;
2115        ldst->datalo_reg = datalo;
2116        ldst->datahi_reg = datahi;
2117        ldst->raddr = tcg_splitwx_to_rx(s->code_ptr);
2118    }
2119}
2120
2121static void tcg_out_exit_tb(TCGContext *s, uintptr_t a0)
2122{
2123    /* Reuse the zeroing that exists for goto_ptr.  */
2124    if (a0 == 0) {
2125        tgen_gotoi(s, S390_CC_ALWAYS, tcg_code_gen_epilogue);
2126    } else {
2127        tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R2, a0);
2128        tgen_gotoi(s, S390_CC_ALWAYS, tb_ret_addr);
2129    }
2130}
2131
2132static void tcg_out_goto_tb(TCGContext *s, int which)
2133{
2134    /*
2135     * Branch displacement must be aligned for atomic patching;
2136     * see if we need to add extra nop before branch
2137     */
2138    if (!QEMU_PTR_IS_ALIGNED(s->code_ptr + 1, 4)) {
2139        tcg_out16(s, NOP);
2140    }
2141    tcg_out16(s, RIL_BRCL | (S390_CC_ALWAYS << 4));
2142    set_jmp_insn_offset(s, which);
2143    s->code_ptr += 2;
2144    set_jmp_reset_offset(s, which);
2145}
2146
2147void tb_target_set_jmp_target(const TranslationBlock *tb, int n,
2148                              uintptr_t jmp_rx, uintptr_t jmp_rw)
2149{
2150    if (!HAVE_FACILITY(GEN_INST_EXT)) {
2151        return;
2152    }
2153    /* patch the branch destination */
2154    uintptr_t addr = tb->jmp_target_addr[n];
2155    intptr_t disp = addr - (jmp_rx - 2);
2156    qatomic_set((int32_t *)jmp_rw, disp / 2);
2157    /* no need to flush icache explicitly */
2158}
2159
2160
2161static void tgen_add(TCGContext *s, TCGType type,
2162                     TCGReg a0, TCGReg a1, TCGReg a2)
2163{
2164    if (a0 != a1) {
2165        tcg_out_insn(s, RX, LA, a0, a1, a2, 0);
2166    } else if (type == TCG_TYPE_I32) {
2167        tcg_out_insn(s, RR, AR, a0, a2);
2168    } else {
2169        tcg_out_insn(s, RRE, AGR, a0, a2);
2170    }
2171}
2172
2173static void tgen_addi(TCGContext *s, TCGType type,
2174                      TCGReg a0, TCGReg a1, tcg_target_long a2)
2175{
2176    if (a0 == a1) {
2177        if (type == TCG_TYPE_I32) {
2178            if (a2 == (int16_t)a2) {
2179                tcg_out_insn(s, RI, AHI, a0, a2);
2180            } else {
2181                tcg_out_insn(s, RIL, AFI, a0, a2);
2182            }
2183            return;
2184        }
2185        if (a2 == (int16_t)a2) {
2186            tcg_out_insn(s, RI, AGHI, a0, a2);
2187            return;
2188        }
2189        if (a2 == (int32_t)a2) {
2190            tcg_out_insn(s, RIL, AGFI, a0, a2);
2191            return;
2192        }
2193        if (a2 == (uint32_t)a2) {
2194            tcg_out_insn(s, RIL, ALGFI, a0, a2);
2195            return;
2196        }
2197        if (-a2 == (uint32_t)-a2) {
2198            tcg_out_insn(s, RIL, SLGFI, a0, -a2);
2199            return;
2200        }
2201    }
2202    tcg_out_mem(s, RX_LA, RXY_LAY, a0, a1, TCG_REG_NONE, a2);
2203}
2204
2205static const TCGOutOpBinary outop_add = {
2206    .base.static_constraint = C_O1_I2(r, r, ri),
2207    .out_rrr = tgen_add,
2208    .out_rri = tgen_addi,
2209};
2210
2211static void tgen_and(TCGContext *s, TCGType type,
2212                     TCGReg a0, TCGReg a1, TCGReg a2)
2213{
2214    if (type != TCG_TYPE_I32) {
2215        tcg_out_insn(s, RRFa, NGRK, a0, a1, a2);
2216    } else if (a0 == a1) {
2217        tcg_out_insn(s, RR, NR, a0, a2);
2218    } else {
2219        tcg_out_insn(s, RRFa, NRK, a0, a1, a2);
2220    }
2221}
2222
2223static void tgen_andi_3(TCGContext *s, TCGType type,
2224                        TCGReg a0, TCGReg a1, tcg_target_long a2)
2225{
2226    tcg_out_mov(s, type, a0, a1);
2227    tgen_andi(s, type, a0, a2);
2228}
2229
2230static const TCGOutOpBinary outop_and = {
2231    .base.static_constraint = C_O1_I2(r, r, rNKR),
2232    .out_rrr = tgen_and,
2233    .out_rri = tgen_andi_3,
2234};
2235
2236static void tgen_andc(TCGContext *s, TCGType type,
2237                      TCGReg a0, TCGReg a1, TCGReg a2)
2238{
2239    if (type == TCG_TYPE_I32) {
2240        tcg_out_insn(s, RRFa, NCRK, a0, a1, a2);
2241    } else {
2242        tcg_out_insn(s, RRFa, NCGRK, a0, a1, a2);
2243    }
2244}
2245
2246static TCGConstraintSetIndex cset_misc3_rrr(TCGType type, unsigned flags)
2247{
2248    return HAVE_FACILITY(MISC_INSN_EXT3) ? C_O1_I2(r, r, r) : C_NotImplemented;
2249}
2250
2251static const TCGOutOpBinary outop_andc = {
2252    .base.static_constraint = C_Dynamic,
2253    .base.dynamic_constraint = cset_misc3_rrr,
2254    .out_rrr = tgen_andc,
2255};
2256
2257static void tgen_clz_int(TCGContext *s, TCGReg dest, TCGReg a1,
2258                         TCGArg a2, int a2const)
2259{
2260    /*
2261     * Since this sets both R and R+1, we have no choice but to store the
2262     * result into R0, allowing R1 == TCG_TMP0 to be clobbered as well.
2263     */
2264    QEMU_BUILD_BUG_ON(TCG_TMP0 != TCG_REG_R1);
2265    tcg_out_insn(s, RRE, FLOGR, TCG_REG_R0, a1);
2266
2267    if (a2const && a2 == 64) {
2268        tcg_out_mov(s, TCG_TYPE_I64, dest, TCG_REG_R0);
2269        return;
2270    }
2271
2272    /*
2273     * Conditions from FLOGR are:
2274     *   2 -> one bit found
2275     *   8 -> no one bit found
2276     */
2277    tgen_movcond_int(s, TCG_TYPE_I64, dest, a2, a2const, TCG_REG_R0, 8, 2);
2278}
2279
2280static void tgen_clz(TCGContext *s, TCGType type,
2281                     TCGReg a0, TCGReg a1, TCGReg a2)
2282{
2283    tgen_clz_int(s, a0, a1, a2, false);
2284}
2285
2286static void tgen_clzi(TCGContext *s, TCGType type,
2287                      TCGReg a0, TCGReg a1, tcg_target_long a2)
2288{
2289    tgen_clz_int(s, a0, a1, a2, true);
2290}
2291
2292static TCGConstraintSetIndex cset_clz(TCGType type, unsigned flags)
2293{
2294    return type == TCG_TYPE_I64 ? C_O1_I2(r, r, rI) : C_NotImplemented;
2295}
2296
2297static const TCGOutOpBinary outop_clz = {
2298    .base.static_constraint = C_Dynamic,
2299    .base.dynamic_constraint = cset_clz,
2300    .out_rrr = tgen_clz,
2301    .out_rri = tgen_clzi,
2302};
2303
2304static void tgen_ctpop(TCGContext *s, TCGType type, TCGReg dest, TCGReg src)
2305{
2306    /* With MIE3, and bit 0 of m4 set, we get the complete result. */
2307    if (HAVE_FACILITY(MISC_INSN_EXT3)) {
2308        if (type == TCG_TYPE_I32) {
2309            tcg_out_ext32u(s, dest, src);
2310            src = dest;
2311        }
2312        tcg_out_insn(s, RRFc, POPCNT, dest, src, 8);
2313        return;
2314    }
2315
2316    /* Without MIE3, each byte gets the count of bits for the byte. */
2317    tcg_out_insn(s, RRFc, POPCNT, dest, src, 0);
2318
2319    /* Multiply to sum each byte at the top of the word. */
2320    if (type == TCG_TYPE_I32) {
2321        tcg_out_insn(s, RIL, MSFI, dest, 0x01010101);
2322        tcg_out_sh32(s, RS_SRL, dest, TCG_REG_NONE, 24);
2323    } else {
2324        tcg_out_movi(s, TCG_TYPE_I64, TCG_TMP0, 0x0101010101010101ull);
2325        tcg_out_insn(s, RRE, MSGR, dest, TCG_TMP0);
2326        tcg_out_sh64(s, RSY_SRLG, dest, dest, TCG_REG_NONE, 56);
2327    }
2328}
2329
2330static const TCGOutOpUnary outop_ctpop = {
2331    .base.static_constraint = C_O1_I1(r, r),
2332    .out_rr = tgen_ctpop,
2333};
2334
2335static const TCGOutOpBinary outop_ctz = {
2336    .base.static_constraint = C_NotImplemented,
2337};
2338
2339static const TCGOutOpBinary outop_divs = {
2340    .base.static_constraint = C_NotImplemented,
2341};
2342
2343static void tgen_divs2(TCGContext *s, TCGType type,
2344                       TCGReg a0, TCGReg a1, TCGReg a4)
2345{
2346    tcg_debug_assert((a1 & 1) == 0);
2347    tcg_debug_assert(a0 == a1 + 1);
2348    if (type == TCG_TYPE_I32) {
2349        tcg_out_insn(s, RR, DR, a1, a4);
2350    } else {
2351        /*
2352         * TODO: Move the sign-extend of the numerator from a2 into a3
2353         * into the tcg backend, instead of in early expansion.  It is
2354         * required for 32-bit DR, but not 64-bit DSGR.
2355         */
2356        tcg_out_insn(s, RRE, DSGR, a1, a4);
2357    }
2358}
2359
2360static const TCGOutOpDivRem outop_divs2 = {
2361    .base.static_constraint = C_O2_I3(o, m, 0, 1, r),
2362    .out_rr01r = tgen_divs2,
2363};
2364
2365static const TCGOutOpBinary outop_divu = {
2366    .base.static_constraint = C_NotImplemented,
2367};
2368
2369static void tgen_divu2(TCGContext *s, TCGType type,
2370                       TCGReg a0, TCGReg a1, TCGReg a4)
2371{
2372    tcg_debug_assert((a1 & 1) == 0);
2373    tcg_debug_assert(a0 == a1 + 1);
2374    if (type == TCG_TYPE_I32) {
2375        tcg_out_insn(s, RRE, DLR, a1, a4);
2376    } else {
2377        tcg_out_insn(s, RRE, DLGR, a1, a4);
2378    }
2379}
2380
2381static const TCGOutOpDivRem outop_divu2 = {
2382    .base.static_constraint = C_O2_I3(o, m, 0, 1, r),
2383    .out_rr01r = tgen_divu2,
2384};
2385
2386static void tgen_eqv(TCGContext *s, TCGType type,
2387                      TCGReg a0, TCGReg a1, TCGReg a2)
2388{
2389    if (type == TCG_TYPE_I32) {
2390        tcg_out_insn(s, RRFa, NXRK, a0, a1, a2);
2391    } else {
2392        tcg_out_insn(s, RRFa, NXGRK, a0, a1, a2);
2393    }
2394}
2395
2396static const TCGOutOpBinary outop_eqv = {
2397    .base.static_constraint = C_Dynamic,
2398    .base.dynamic_constraint = cset_misc3_rrr,
2399    .out_rrr = tgen_eqv,
2400};
2401
2402static void tgen_mul(TCGContext *s, TCGType type,
2403                     TCGReg a0, TCGReg a1, TCGReg a2)
2404{
2405    if (type == TCG_TYPE_I32) {
2406        if (a0 == a1) {
2407            tcg_out_insn(s, RRE, MSR, a0, a2);
2408        } else {
2409            tcg_out_insn(s, RRFa, MSRKC, a0, a1, a2);
2410        }
2411    } else {
2412        if (a0 == a1) {
2413            tcg_out_insn(s, RRE, MSGR, a0, a2);
2414        } else {
2415            tcg_out_insn(s, RRFa, MSGRKC, a0, a1, a2);
2416        }
2417    }
2418}
2419
2420static void tgen_muli(TCGContext *s, TCGType type,
2421                      TCGReg a0, TCGReg a1, tcg_target_long a2)
2422{
2423    tcg_out_mov(s, type, a0, a1);
2424    if (type == TCG_TYPE_I32) {
2425        if (a2 == (int16_t)a2) {
2426            tcg_out_insn(s, RI, MHI, a0, a2);
2427        } else {
2428            tcg_out_insn(s, RIL, MSFI, a0, a2);
2429        }
2430    } else {
2431        if (a2 == (int16_t)a2) {
2432            tcg_out_insn(s, RI, MGHI, a0, a2);
2433        } else {
2434            tcg_out_insn(s, RIL, MSGFI, a0, a2);
2435        }
2436    }
2437}
2438
2439static TCGConstraintSetIndex cset_mul(TCGType type, unsigned flags)
2440{
2441    return (HAVE_FACILITY(MISC_INSN_EXT2)
2442            ? C_O1_I2(r, r, rJ)
2443            : C_O1_I2(r, 0, rJ));
2444}
2445
2446static const TCGOutOpBinary outop_mul = {
2447    .base.static_constraint = C_Dynamic,
2448    .base.dynamic_constraint = cset_mul,
2449    .out_rrr = tgen_mul,
2450    .out_rri = tgen_muli,
2451};
2452
2453static void tgen_muls2(TCGContext *s, TCGType type,
2454                       TCGReg a0, TCGReg a1, TCGReg a2, TCGReg a3)
2455{
2456    tcg_debug_assert((a1 & 1) == 0);
2457    tcg_debug_assert(a0 == a1 + 1);
2458    tcg_out_insn(s, RRFa, MGRK, a1, a2, a3);
2459}
2460
2461static TCGConstraintSetIndex cset_muls2(TCGType type, unsigned flags)
2462{
2463    return (type == TCG_TYPE_I64 && HAVE_FACILITY(MISC_INSN_EXT2)
2464            ? C_O2_I2(o, m, r, r) : C_NotImplemented);
2465}
2466
2467static const TCGOutOpMul2 outop_muls2 = {
2468    .base.static_constraint = C_Dynamic,
2469    .base.dynamic_constraint = cset_muls2,
2470    .out_rrrr = tgen_muls2,
2471};
2472
2473static const TCGOutOpBinary outop_mulsh = {
2474    .base.static_constraint = C_NotImplemented,
2475};
2476
2477static void tgen_mulu2(TCGContext *s, TCGType type,
2478                       TCGReg a0, TCGReg a1, TCGReg a2, TCGReg a3)
2479{
2480    tcg_debug_assert(a0 == a2);
2481    tcg_debug_assert((a1 & 1) == 0);
2482    tcg_debug_assert(a0 == a1 + 1);
2483    tcg_out_insn(s, RRE, MLGR, a1, a3);
2484}
2485
2486static TCGConstraintSetIndex cset_mulu2(TCGType type, unsigned flags)
2487{
2488    return (type == TCG_TYPE_I64 && HAVE_FACILITY(MISC_INSN_EXT2)
2489            ? C_O2_I2(o, m, 0, r) : C_NotImplemented);
2490}
2491
2492static const TCGOutOpMul2 outop_mulu2 = {
2493    .base.static_constraint = C_Dynamic,
2494    .base.dynamic_constraint = cset_mulu2,
2495    .out_rrrr = tgen_mulu2,
2496};
2497
2498static const TCGOutOpBinary outop_muluh = {
2499    .base.static_constraint = C_NotImplemented,
2500};
2501
2502static void tgen_nand(TCGContext *s, TCGType type,
2503                      TCGReg a0, TCGReg a1, TCGReg a2)
2504{
2505    if (type == TCG_TYPE_I32) {
2506        tcg_out_insn(s, RRFa, NNRK, a0, a1, a2);
2507    } else {
2508        tcg_out_insn(s, RRFa, NNGRK, a0, a1, a2);
2509    }
2510}
2511
2512static const TCGOutOpBinary outop_nand = {
2513    .base.static_constraint = C_Dynamic,
2514    .base.dynamic_constraint = cset_misc3_rrr,
2515    .out_rrr = tgen_nand,
2516};
2517
2518static void tgen_nor(TCGContext *s, TCGType type,
2519                      TCGReg a0, TCGReg a1, TCGReg a2)
2520{
2521    if (type == TCG_TYPE_I32) {
2522        tcg_out_insn(s, RRFa, NORK, a0, a1, a2);
2523    } else {
2524        tcg_out_insn(s, RRFa, NOGRK, a0, a1, a2);
2525    }
2526}
2527
2528static const TCGOutOpBinary outop_nor = {
2529    .base.static_constraint = C_Dynamic,
2530    .base.dynamic_constraint = cset_misc3_rrr,
2531    .out_rrr = tgen_nor,
2532};
2533
2534static void tgen_or(TCGContext *s, TCGType type,
2535                     TCGReg a0, TCGReg a1, TCGReg a2)
2536{
2537    if (type != TCG_TYPE_I32) {
2538        tcg_out_insn(s, RRFa, OGRK, a0, a1, a2);
2539    } else if (a0 == a1) {
2540        tcg_out_insn(s, RR, OR, a0, a2);
2541    } else {
2542        tcg_out_insn(s, RRFa, ORK, a0, a1, a2);
2543    }
2544}
2545
2546static void tgen_ori_3(TCGContext *s, TCGType type,
2547                        TCGReg a0, TCGReg a1, tcg_target_long a2)
2548{
2549    tcg_out_mov(s, type, a0, a1);
2550    tgen_ori(s, a0, type == TCG_TYPE_I32 ? (uint32_t)a2 : a2);
2551}
2552
2553static const TCGOutOpBinary outop_or = {
2554    .base.static_constraint = C_O1_I2(r, r, rK),
2555    .out_rrr = tgen_or,
2556    .out_rri = tgen_ori_3,
2557};
2558
2559static void tgen_orc(TCGContext *s, TCGType type,
2560                     TCGReg a0, TCGReg a1, TCGReg a2)
2561{
2562    if (type == TCG_TYPE_I32) {
2563        tcg_out_insn(s, RRFa, OCRK, a0, a1, a2);
2564    } else {
2565        tcg_out_insn(s, RRFa, OCGRK, a0, a1, a2);
2566    }
2567}
2568
2569static const TCGOutOpBinary outop_orc = {
2570    .base.static_constraint = C_Dynamic,
2571    .base.dynamic_constraint = cset_misc3_rrr,
2572    .out_rrr = tgen_orc,
2573};
2574
2575static const TCGOutOpBinary outop_rems = {
2576    .base.static_constraint = C_NotImplemented,
2577};
2578
2579static const TCGOutOpBinary outop_remu = {
2580    .base.static_constraint = C_NotImplemented,
2581};
2582
2583static void tgen_rotl_int(TCGContext *s, TCGType type, TCGReg dst,
2584                          TCGReg src, TCGReg v, tcg_target_long i)
2585{
2586    S390Opcode insn = type == TCG_TYPE_I32 ? RSY_RLL : RSY_RLLG;
2587    tcg_out_sh64(s, insn, dst, src, v, i);
2588}
2589
2590static void tgen_rotl(TCGContext *s, TCGType type,
2591                      TCGReg a0, TCGReg a1, TCGReg a2)
2592{
2593    tgen_rotl_int(s, type, a0, a1, a2, 0);
2594}
2595
2596static void tgen_rotli(TCGContext *s, TCGType type,
2597                       TCGReg a0, TCGReg a1, tcg_target_long a2)
2598{
2599    tgen_rotl_int(s, type, a0, a1, TCG_REG_NONE, a2);
2600}
2601
2602static const TCGOutOpBinary outop_rotl = {
2603    .base.static_constraint = C_O1_I2(r, r, ri),
2604    .out_rrr = tgen_rotl,
2605    .out_rri = tgen_rotli,
2606};
2607
2608static const TCGOutOpBinary outop_rotr = {
2609    .base.static_constraint = C_NotImplemented,
2610};
2611
2612static void tgen_sar_int(TCGContext *s, TCGType type, TCGReg dst,
2613                         TCGReg src, TCGReg v, tcg_target_long i)
2614{
2615    if (type != TCG_TYPE_I32) {
2616        tcg_out_sh64(s, RSY_SRAG, dst, src, v, i);
2617    } else if (dst == src) {
2618        tcg_out_sh32(s, RS_SRA, dst, v, i);
2619    } else {
2620        tcg_out_sh64(s, RSY_SRAK, dst, src, v, i);
2621    }
2622}
2623
2624static void tgen_sar(TCGContext *s, TCGType type,
2625                     TCGReg a0, TCGReg a1, TCGReg a2)
2626{
2627    tgen_sar_int(s, type, a0, a1, a2, 0);
2628}
2629
2630static void tgen_sari(TCGContext *s, TCGType type,
2631                      TCGReg a0, TCGReg a1, tcg_target_long a2)
2632{
2633    tgen_sar_int(s, type, a0, a1, TCG_REG_NONE, a2);
2634}
2635
2636static const TCGOutOpBinary outop_sar = {
2637    .base.static_constraint = C_O1_I2(r, r, ri),
2638    .out_rrr = tgen_sar,
2639    .out_rri = tgen_sari,
2640};
2641
2642static void tgen_shl_int(TCGContext *s, TCGType type, TCGReg dst,
2643                         TCGReg src, TCGReg v, tcg_target_long i)
2644{
2645    if (type != TCG_TYPE_I32) {
2646        tcg_out_sh64(s, RSY_SLLG, dst, src, v, i);
2647    } else if (dst == src) {
2648        tcg_out_sh32(s, RS_SLL, dst, v, i);
2649    } else {
2650        tcg_out_sh64(s, RSY_SLLK, dst, src, v, i);
2651    }
2652}
2653
2654static void tgen_shl(TCGContext *s, TCGType type,
2655                     TCGReg a0, TCGReg a1, TCGReg a2)
2656{
2657    tgen_shl_int(s, type, a0, a1, a2, 0);
2658}
2659
2660static void tgen_shli(TCGContext *s, TCGType type,
2661                      TCGReg a0, TCGReg a1, tcg_target_long a2)
2662{
2663    tgen_shl_int(s, type, a0, a1, TCG_REG_NONE, a2);
2664}
2665
2666static const TCGOutOpBinary outop_shl = {
2667    .base.static_constraint = C_O1_I2(r, r, ri),
2668    .out_rrr = tgen_shl,
2669    .out_rri = tgen_shli,
2670};
2671
2672static void tgen_shr_int(TCGContext *s, TCGType type, TCGReg dst,
2673                         TCGReg src, TCGReg v, tcg_target_long i)
2674{
2675    if (type != TCG_TYPE_I32) {
2676        tcg_out_sh64(s, RSY_SRLG, dst, src, v, i);
2677    } else if (dst == src) {
2678        tcg_out_sh32(s, RS_SRL, dst, v, i);
2679    } else {
2680        tcg_out_sh64(s, RSY_SRLK, dst, src, v, i);
2681    }
2682}
2683
2684static void tgen_shr(TCGContext *s, TCGType type,
2685                     TCGReg a0, TCGReg a1, TCGReg a2)
2686{
2687    tgen_shr_int(s, type, a0, a1, a2, 0);
2688}
2689
2690static void tgen_shri(TCGContext *s, TCGType type,
2691                      TCGReg a0, TCGReg a1, tcg_target_long a2)
2692{
2693    tgen_shr_int(s, type, a0, a1, TCG_REG_NONE, a2);
2694}
2695
2696static const TCGOutOpBinary outop_shr = {
2697    .base.static_constraint = C_O1_I2(r, r, ri),
2698    .out_rrr = tgen_shr,
2699    .out_rri = tgen_shri,
2700};
2701
2702static void tgen_sub(TCGContext *s, TCGType type,
2703                     TCGReg a0, TCGReg a1, TCGReg a2)
2704{
2705    if (type != TCG_TYPE_I32) {
2706        tcg_out_insn(s, RRFa, SGRK, a0, a1, a2);
2707    } else if (a0 == a1) {
2708        tcg_out_insn(s, RR, SR, a0, a2);
2709    } else {
2710        tcg_out_insn(s, RRFa, SRK, a0, a1, a2);
2711    }
2712}
2713
2714static const TCGOutOpSubtract outop_sub = {
2715    .base.static_constraint = C_O1_I2(r, r, r),
2716    .out_rrr = tgen_sub,
2717};
2718
2719static void tgen_xor(TCGContext *s, TCGType type,
2720                     TCGReg a0, TCGReg a1, TCGReg a2)
2721{
2722    if (type != TCG_TYPE_I32) {
2723        tcg_out_insn(s, RRFa, XGRK, a0, a1, a2);
2724    } else if (a0 == a1) {
2725        tcg_out_insn(s, RR, XR, a0, a2);
2726    } else {
2727        tcg_out_insn(s, RRFa, XRK, a0, a1, a2);
2728    }
2729}
2730
2731static void tgen_xori_3(TCGContext *s, TCGType type,
2732                        TCGReg a0, TCGReg a1, tcg_target_long a2)
2733{
2734    tcg_out_mov(s, type, a0, a1);
2735    tgen_xori(s, a0, type == TCG_TYPE_I32 ? (uint32_t)a2 : a2);
2736}
2737
2738static const TCGOutOpBinary outop_xor = {
2739    .base.static_constraint = C_O1_I2(r, r, rK),
2740    .out_rrr = tgen_xor,
2741    .out_rri = tgen_xori_3,
2742};
2743
2744static void tgen_bswap16(TCGContext *s, TCGType type,
2745                         TCGReg a0, TCGReg a1, unsigned flags)
2746{
2747    if (type == TCG_TYPE_I32) {
2748        tcg_out_insn(s, RRE, LRVR, a0, a1);
2749        tcg_out_sh32(s, (flags & TCG_BSWAP_OS ? RS_SRA : RS_SRL),
2750                     a0, TCG_REG_NONE, 16);
2751    } else {
2752        tcg_out_insn(s, RRE, LRVGR, a0, a1);
2753        tcg_out_sh64(s, (flags & TCG_BSWAP_OS ? RSY_SRAG : RSY_SRLG),
2754                     a0, a0, TCG_REG_NONE, 48);
2755    }
2756}
2757
2758static const TCGOutOpBswap outop_bswap16 = {
2759    .base.static_constraint = C_O1_I1(r, r),
2760    .out_rr = tgen_bswap16,
2761};
2762
2763static void tgen_bswap32(TCGContext *s, TCGType type,
2764                         TCGReg a0, TCGReg a1, unsigned flags)
2765{
2766    tcg_out_insn(s, RRE, LRVR, a0, a1);
2767    if (flags & TCG_BSWAP_OS) {
2768        tcg_out_ext32s(s, a0, a0);
2769    } else if ((flags & (TCG_BSWAP_IZ | TCG_BSWAP_OZ)) == TCG_BSWAP_OZ) {
2770        tcg_out_ext32u(s, a0, a0);
2771    }
2772}
2773
2774static const TCGOutOpBswap outop_bswap32 = {
2775    .base.static_constraint = C_O1_I1(r, r),
2776    .out_rr = tgen_bswap32,
2777};
2778
2779static void tgen_bswap64(TCGContext *s, TCGType type, TCGReg a0, TCGReg a1)
2780{
2781    tcg_out_insn(s, RRE, LRVGR, a0, a1);
2782}
2783
2784static const TCGOutOpUnary outop_bswap64 = {
2785    .base.static_constraint = C_O1_I1(r, r),
2786    .out_rr = tgen_bswap64,
2787};
2788
2789static void tgen_neg(TCGContext *s, TCGType type, TCGReg a0, TCGReg a1)
2790{
2791    if (type == TCG_TYPE_I32) {
2792        tcg_out_insn(s, RR, LCR, a0, a1);
2793    } else {
2794        tcg_out_insn(s, RRE, LCGR, a0, a1);
2795    }
2796}
2797
2798static const TCGOutOpUnary outop_neg = {
2799    .base.static_constraint = C_O1_I1(r, r),
2800    .out_rr = tgen_neg,
2801};
2802
2803static void tgen_not(TCGContext *s, TCGType type, TCGReg a0, TCGReg a1)
2804{
2805    tgen_nor(s, type, a0, a1, a1);
2806}
2807
2808static TCGConstraintSetIndex cset_not(TCGType type, unsigned flags)
2809{
2810    return HAVE_FACILITY(MISC_INSN_EXT3) ? C_O1_I1(r, r) : C_NotImplemented;
2811}
2812
2813static const TCGOutOpUnary outop_not = {
2814    .base.static_constraint = C_Dynamic,
2815    .base.dynamic_constraint = cset_not,
2816    .out_rr = tgen_not,
2817};
2818
2819
2820# define OP_32_64(x) \
2821        case glue(glue(INDEX_op_,x),_i32): \
2822        case glue(glue(INDEX_op_,x),_i64)
2823
2824static void tcg_out_op(TCGContext *s, TCGOpcode opc, TCGType type,
2825                       const TCGArg args[TCG_MAX_OP_ARGS],
2826                       const int const_args[TCG_MAX_OP_ARGS])
2827{
2828    TCGArg a0, a1, a2;
2829
2830    switch (opc) {
2831    case INDEX_op_goto_ptr:
2832        a0 = args[0];
2833        tcg_out_insn(s, RR, BCR, S390_CC_ALWAYS, a0);
2834        break;
2835
2836    OP_32_64(ld8u):
2837        /* ??? LLC (RXY format) is only present with the extended-immediate
2838           facility, whereas LLGC is always present.  */
2839        tcg_out_mem(s, 0, RXY_LLGC, args[0], args[1], TCG_REG_NONE, args[2]);
2840        break;
2841
2842    OP_32_64(ld8s):
2843        /* ??? LB is no smaller than LGB, so no point to using it.  */
2844        tcg_out_mem(s, 0, RXY_LGB, args[0], args[1], TCG_REG_NONE, args[2]);
2845        break;
2846
2847    OP_32_64(ld16u):
2848        /* ??? LLH (RXY format) is only present with the extended-immediate
2849           facility, whereas LLGH is always present.  */
2850        tcg_out_mem(s, 0, RXY_LLGH, args[0], args[1], TCG_REG_NONE, args[2]);
2851        break;
2852
2853    case INDEX_op_ld16s_i32:
2854        tcg_out_mem(s, RX_LH, RXY_LHY, args[0], args[1], TCG_REG_NONE, args[2]);
2855        break;
2856
2857    case INDEX_op_ld_i32:
2858        tcg_out_ld(s, TCG_TYPE_I32, args[0], args[1], args[2]);
2859        break;
2860
2861    OP_32_64(st8):
2862        tcg_out_mem(s, RX_STC, RXY_STCY, args[0], args[1],
2863                    TCG_REG_NONE, args[2]);
2864        break;
2865
2866    OP_32_64(st16):
2867        tcg_out_mem(s, RX_STH, RXY_STHY, args[0], args[1],
2868                    TCG_REG_NONE, args[2]);
2869        break;
2870
2871    case INDEX_op_st_i32:
2872        tcg_out_st(s, TCG_TYPE_I32, args[0], args[1], args[2]);
2873        break;
2874
2875    case INDEX_op_add2_i32:
2876        if (const_args[4]) {
2877            tcg_out_insn(s, RIL, ALFI, args[0], args[4]);
2878        } else {
2879            tcg_out_insn(s, RR, ALR, args[0], args[4]);
2880        }
2881        tcg_out_insn(s, RRE, ALCR, args[1], args[5]);
2882        break;
2883    case INDEX_op_sub2_i32:
2884        if (const_args[4]) {
2885            tcg_out_insn(s, RIL, SLFI, args[0], args[4]);
2886        } else {
2887            tcg_out_insn(s, RR, SLR, args[0], args[4]);
2888        }
2889        tcg_out_insn(s, RRE, SLBR, args[1], args[5]);
2890        break;
2891
2892    case INDEX_op_br:
2893        tgen_branch(s, S390_CC_ALWAYS, arg_label(args[0]));
2894        break;
2895
2896    case INDEX_op_qemu_ld_i32:
2897        tcg_out_qemu_ld(s, args[0], args[1], args[2], TCG_TYPE_I32);
2898        break;
2899    case INDEX_op_qemu_ld_i64:
2900        tcg_out_qemu_ld(s, args[0], args[1], args[2], TCG_TYPE_I64);
2901        break;
2902    case INDEX_op_qemu_st_i32:
2903        tcg_out_qemu_st(s, args[0], args[1], args[2], TCG_TYPE_I32);
2904        break;
2905    case INDEX_op_qemu_st_i64:
2906        tcg_out_qemu_st(s, args[0], args[1], args[2], TCG_TYPE_I64);
2907        break;
2908    case INDEX_op_qemu_ld_i128:
2909        tcg_out_qemu_ldst_i128(s, args[0], args[1], args[2], args[3], true);
2910        break;
2911    case INDEX_op_qemu_st_i128:
2912        tcg_out_qemu_ldst_i128(s, args[0], args[1], args[2], args[3], false);
2913        break;
2914
2915    case INDEX_op_ld16s_i64:
2916        tcg_out_mem(s, 0, RXY_LGH, args[0], args[1], TCG_REG_NONE, args[2]);
2917        break;
2918    case INDEX_op_ld32u_i64:
2919        tcg_out_mem(s, 0, RXY_LLGF, args[0], args[1], TCG_REG_NONE, args[2]);
2920        break;
2921    case INDEX_op_ld32s_i64:
2922        tcg_out_mem(s, 0, RXY_LGF, args[0], args[1], TCG_REG_NONE, args[2]);
2923        break;
2924    case INDEX_op_ld_i64:
2925        tcg_out_ld(s, TCG_TYPE_I64, args[0], args[1], args[2]);
2926        break;
2927
2928    case INDEX_op_st32_i64:
2929        tcg_out_st(s, TCG_TYPE_I32, args[0], args[1], args[2]);
2930        break;
2931    case INDEX_op_st_i64:
2932        tcg_out_st(s, TCG_TYPE_I64, args[0], args[1], args[2]);
2933        break;
2934
2935    case INDEX_op_add2_i64:
2936        if (const_args[4]) {
2937            if ((int64_t)args[4] >= 0) {
2938                tcg_out_insn(s, RIL, ALGFI, args[0], args[4]);
2939            } else {
2940                tcg_out_insn(s, RIL, SLGFI, args[0], -args[4]);
2941            }
2942        } else {
2943            tcg_out_insn(s, RRE, ALGR, args[0], args[4]);
2944        }
2945        tcg_out_insn(s, RRE, ALCGR, args[1], args[5]);
2946        break;
2947    case INDEX_op_sub2_i64:
2948        if (const_args[4]) {
2949            if ((int64_t)args[4] >= 0) {
2950                tcg_out_insn(s, RIL, SLGFI, args[0], args[4]);
2951            } else {
2952                tcg_out_insn(s, RIL, ALGFI, args[0], -args[4]);
2953            }
2954        } else {
2955            tcg_out_insn(s, RRE, SLGR, args[0], args[4]);
2956        }
2957        tcg_out_insn(s, RRE, SLBGR, args[1], args[5]);
2958        break;
2959
2960    OP_32_64(deposit):
2961        a0 = args[0], a1 = args[1], a2 = args[2];
2962        if (const_args[1]) {
2963            tgen_deposit(s, a0, a2, args[3], args[4], 1);
2964        } else {
2965            /* Since we can't support "0Z" as a constraint, we allow a1 in
2966               any register.  Fix things up as if a matching constraint.  */
2967            if (a0 != a1) {
2968                if (a0 == a2) {
2969                    tcg_out_mov(s, type, TCG_TMP0, a2);
2970                    a2 = TCG_TMP0;
2971                }
2972                tcg_out_mov(s, type, a0, a1);
2973            }
2974            tgen_deposit(s, a0, a2, args[3], args[4], 0);
2975        }
2976        break;
2977
2978    OP_32_64(extract):
2979        tgen_extract(s, args[0], args[1], args[2], args[3]);
2980        break;
2981    OP_32_64(sextract):
2982        tgen_sextract(s, args[0], args[1], args[2], args[3]);
2983        break;
2984
2985    case INDEX_op_mb:
2986        /* The host memory model is quite strong, we simply need to
2987           serialize the instruction stream.  */
2988        if (args[0] & TCG_MO_ST_LD) {
2989            /* fast-bcr-serialization facility (45) is present */
2990            tcg_out_insn(s, RR, BCR, 14, 0);
2991        }
2992        break;
2993
2994    case INDEX_op_call:     /* Always emitted via tcg_out_call.  */
2995    case INDEX_op_exit_tb:  /* Always emitted via tcg_out_exit_tb.  */
2996    case INDEX_op_goto_tb:  /* Always emitted via tcg_out_goto_tb.  */
2997    case INDEX_op_ext_i32_i64:  /* Always emitted via tcg_reg_alloc_op.  */
2998    case INDEX_op_extu_i32_i64:
2999    case INDEX_op_extrl_i64_i32:
3000    default:
3001        g_assert_not_reached();
3002    }
3003}
3004
3005static bool tcg_out_dup_vec(TCGContext *s, TCGType type, unsigned vece,
3006                            TCGReg dst, TCGReg src)
3007{
3008    if (is_general_reg(src)) {
3009        /* Replicate general register into two MO_64. */
3010        tcg_out_insn(s, VRRf, VLVGP, dst, src, src);
3011        if (vece == MO_64) {
3012            return true;
3013        }
3014        src = dst;
3015    }
3016
3017    /*
3018     * Recall that the "standard" integer, within a vector, is the
3019     * rightmost element of the leftmost doubleword, a-la VLLEZ.
3020     */
3021    tcg_out_insn(s, VRIc, VREP, dst, (8 >> vece) - 1, src, vece);
3022    return true;
3023}
3024
3025static bool tcg_out_dupm_vec(TCGContext *s, TCGType type, unsigned vece,
3026                             TCGReg dst, TCGReg base, intptr_t offset)
3027{
3028    tcg_out_vrx_mem(s, VRX_VLREP, dst, base, TCG_REG_NONE, offset, vece);
3029    return true;
3030}
3031
3032static void tcg_out_dupi_vec(TCGContext *s, TCGType type, unsigned vece,
3033                             TCGReg dst, int64_t val)
3034{
3035    int i, mask, msb, lsb;
3036
3037    /* Look for int16_t elements.  */
3038    if (vece <= MO_16 ||
3039        (vece == MO_32 ? (int32_t)val : val) == (int16_t)val) {
3040        tcg_out_insn(s, VRIa, VREPI, dst, val, vece);
3041        return;
3042    }
3043
3044    /* Look for bit masks.  */
3045    if (vece == MO_32) {
3046        if (risbg_mask((int32_t)val)) {
3047            /* Handle wraparound by swapping msb and lsb.  */
3048            if ((val & 0x80000001u) == 0x80000001u) {
3049                msb = 32 - ctz32(~val);
3050                lsb = clz32(~val) - 1;
3051            } else {
3052                msb = clz32(val);
3053                lsb = 31 - ctz32(val);
3054            }
3055            tcg_out_insn(s, VRIb, VGM, dst, msb, lsb, MO_32);
3056            return;
3057        }
3058    } else {
3059        if (risbg_mask(val)) {
3060            /* Handle wraparound by swapping msb and lsb.  */
3061            if ((val & 0x8000000000000001ull) == 0x8000000000000001ull) {
3062                /* Handle wraparound by swapping msb and lsb.  */
3063                msb = 64 - ctz64(~val);
3064                lsb = clz64(~val) - 1;
3065            } else {
3066                msb = clz64(val);
3067                lsb = 63 - ctz64(val);
3068            }
3069            tcg_out_insn(s, VRIb, VGM, dst, msb, lsb, MO_64);
3070            return;
3071        }
3072    }
3073
3074    /* Look for all bytes 0x00 or 0xff.  */
3075    for (i = mask = 0; i < 8; i++) {
3076        uint8_t byte = val >> (i * 8);
3077        if (byte == 0xff) {
3078            mask |= 1 << i;
3079        } else if (byte != 0) {
3080            break;
3081        }
3082    }
3083    if (i == 8) {
3084        tcg_out_insn(s, VRIa, VGBM, dst, mask * 0x0101, 0);
3085        return;
3086    }
3087
3088    /* Otherwise, stuff it in the constant pool.  */
3089    tcg_out_insn(s, RIL, LARL, TCG_TMP0, 0);
3090    new_pool_label(s, val, R_390_PC32DBL, s->code_ptr - 2, 2);
3091    tcg_out_insn(s, VRX, VLREP, dst, TCG_TMP0, TCG_REG_NONE, 0, MO_64);
3092}
3093
3094static bool tcg_out_cmp_vec_noinv(TCGContext *s, unsigned vece, TCGReg a0,
3095                                  TCGReg a1, TCGReg a2, TCGCond cond)
3096{
3097    bool need_swap = false, need_inv = false;
3098
3099    switch (cond) {
3100    case TCG_COND_EQ:
3101    case TCG_COND_GT:
3102    case TCG_COND_GTU:
3103        break;
3104    case TCG_COND_NE:
3105    case TCG_COND_LE:
3106    case TCG_COND_LEU:
3107        need_inv = true;
3108        break;
3109    case TCG_COND_LT:
3110    case TCG_COND_LTU:
3111        need_swap = true;
3112        break;
3113    case TCG_COND_GE:
3114    case TCG_COND_GEU:
3115        need_swap = need_inv = true;
3116        break;
3117    default:
3118        g_assert_not_reached();
3119    }
3120
3121    if (need_inv) {
3122        cond = tcg_invert_cond(cond);
3123    }
3124    if (need_swap) {
3125        TCGReg swap = a1;
3126        a1 = a2;
3127        a2 = swap;
3128        cond = tcg_swap_cond(cond);
3129    }
3130
3131    switch (cond) {
3132    case TCG_COND_EQ:
3133        tcg_out_insn(s, VRRc, VCEQ, a0, a1, a2, vece);
3134        break;
3135    case TCG_COND_GT:
3136        tcg_out_insn(s, VRRc, VCH, a0, a1, a2, vece);
3137        break;
3138    case TCG_COND_GTU:
3139        tcg_out_insn(s, VRRc, VCHL, a0, a1, a2, vece);
3140        break;
3141    default:
3142        g_assert_not_reached();
3143    }
3144    return need_inv;
3145}
3146
3147static void tcg_out_cmp_vec(TCGContext *s, unsigned vece, TCGReg a0,
3148                            TCGReg a1, TCGReg a2, TCGCond cond)
3149{
3150    if (tcg_out_cmp_vec_noinv(s, vece, a0, a1, a2, cond)) {
3151        tcg_out_insn(s, VRRc, VNO, a0, a0, a0, 0);
3152    }
3153}
3154
3155static void tcg_out_cmpsel_vec(TCGContext *s, unsigned vece, TCGReg a0,
3156                               TCGReg c1, TCGReg c2, TCGArg v3,
3157                               int const_v3, TCGReg v4, TCGCond cond)
3158{
3159    bool inv = tcg_out_cmp_vec_noinv(s, vece, TCG_VEC_TMP0, c1, c2, cond);
3160
3161    if (!const_v3) {
3162        if (inv) {
3163            tcg_out_insn(s, VRRe, VSEL, a0, v4, v3, TCG_VEC_TMP0);
3164        } else {
3165            tcg_out_insn(s, VRRe, VSEL, a0, v3, v4, TCG_VEC_TMP0);
3166        }
3167    } else if (v3) {
3168        if (inv) {
3169            tcg_out_insn(s, VRRc, VOC, a0, v4, TCG_VEC_TMP0, 0);
3170        } else {
3171            tcg_out_insn(s, VRRc, VO, a0, v4, TCG_VEC_TMP0, 0);
3172        }
3173    } else {
3174        if (inv) {
3175            tcg_out_insn(s, VRRc, VN, a0, v4, TCG_VEC_TMP0, 0);
3176        } else {
3177            tcg_out_insn(s, VRRc, VNC, a0, v4, TCG_VEC_TMP0, 0);
3178        }
3179    }
3180}
3181
3182static void tcg_out_vec_op(TCGContext *s, TCGOpcode opc,
3183                           unsigned vecl, unsigned vece,
3184                           const TCGArg args[TCG_MAX_OP_ARGS],
3185                           const int const_args[TCG_MAX_OP_ARGS])
3186{
3187    TCGType type = vecl + TCG_TYPE_V64;
3188    TCGArg a0 = args[0], a1 = args[1], a2 = args[2];
3189
3190    switch (opc) {
3191    case INDEX_op_ld_vec:
3192        tcg_out_ld(s, type, a0, a1, a2);
3193        break;
3194    case INDEX_op_st_vec:
3195        tcg_out_st(s, type, a0, a1, a2);
3196        break;
3197    case INDEX_op_dupm_vec:
3198        tcg_out_dupm_vec(s, type, vece, a0, a1, a2);
3199        break;
3200
3201    case INDEX_op_abs_vec:
3202        tcg_out_insn(s, VRRa, VLP, a0, a1, vece);
3203        break;
3204    case INDEX_op_neg_vec:
3205        tcg_out_insn(s, VRRa, VLC, a0, a1, vece);
3206        break;
3207    case INDEX_op_not_vec:
3208        tcg_out_insn(s, VRRc, VNO, a0, a1, a1, 0);
3209        break;
3210
3211    case INDEX_op_add_vec:
3212        tcg_out_insn(s, VRRc, VA, a0, a1, a2, vece);
3213        break;
3214    case INDEX_op_sub_vec:
3215        tcg_out_insn(s, VRRc, VS, a0, a1, a2, vece);
3216        break;
3217    case INDEX_op_and_vec:
3218        tcg_out_insn(s, VRRc, VN, a0, a1, a2, 0);
3219        break;
3220    case INDEX_op_andc_vec:
3221        tcg_out_insn(s, VRRc, VNC, a0, a1, a2, 0);
3222        break;
3223    case INDEX_op_mul_vec:
3224        tcg_out_insn(s, VRRc, VML, a0, a1, a2, vece);
3225        break;
3226    case INDEX_op_or_vec:
3227        tcg_out_insn(s, VRRc, VO, a0, a1, a2, 0);
3228        break;
3229    case INDEX_op_orc_vec:
3230        tcg_out_insn(s, VRRc, VOC, a0, a1, a2, 0);
3231        break;
3232    case INDEX_op_xor_vec:
3233        tcg_out_insn(s, VRRc, VX, a0, a1, a2, 0);
3234        break;
3235    case INDEX_op_nand_vec:
3236        tcg_out_insn(s, VRRc, VNN, a0, a1, a2, 0);
3237        break;
3238    case INDEX_op_nor_vec:
3239        tcg_out_insn(s, VRRc, VNO, a0, a1, a2, 0);
3240        break;
3241    case INDEX_op_eqv_vec:
3242        tcg_out_insn(s, VRRc, VNX, a0, a1, a2, 0);
3243        break;
3244
3245    case INDEX_op_shli_vec:
3246        tcg_out_insn(s, VRSa, VESL, a0, a2, TCG_REG_NONE, a1, vece);
3247        break;
3248    case INDEX_op_shri_vec:
3249        tcg_out_insn(s, VRSa, VESRL, a0, a2, TCG_REG_NONE, a1, vece);
3250        break;
3251    case INDEX_op_sari_vec:
3252        tcg_out_insn(s, VRSa, VESRA, a0, a2, TCG_REG_NONE, a1, vece);
3253        break;
3254    case INDEX_op_rotli_vec:
3255        tcg_out_insn(s, VRSa, VERLL, a0, a2, TCG_REG_NONE, a1, vece);
3256        break;
3257    case INDEX_op_shls_vec:
3258        tcg_out_insn(s, VRSa, VESL, a0, 0, a2, a1, vece);
3259        break;
3260    case INDEX_op_shrs_vec:
3261        tcg_out_insn(s, VRSa, VESRL, a0, 0, a2, a1, vece);
3262        break;
3263    case INDEX_op_sars_vec:
3264        tcg_out_insn(s, VRSa, VESRA, a0, 0, a2, a1, vece);
3265        break;
3266    case INDEX_op_rotls_vec:
3267        tcg_out_insn(s, VRSa, VERLL, a0, 0, a2, a1, vece);
3268        break;
3269    case INDEX_op_shlv_vec:
3270        tcg_out_insn(s, VRRc, VESLV, a0, a1, a2, vece);
3271        break;
3272    case INDEX_op_shrv_vec:
3273        tcg_out_insn(s, VRRc, VESRLV, a0, a1, a2, vece);
3274        break;
3275    case INDEX_op_sarv_vec:
3276        tcg_out_insn(s, VRRc, VESRAV, a0, a1, a2, vece);
3277        break;
3278    case INDEX_op_rotlv_vec:
3279        tcg_out_insn(s, VRRc, VERLLV, a0, a1, a2, vece);
3280        break;
3281
3282    case INDEX_op_smin_vec:
3283        tcg_out_insn(s, VRRc, VMN, a0, a1, a2, vece);
3284        break;
3285    case INDEX_op_smax_vec:
3286        tcg_out_insn(s, VRRc, VMX, a0, a1, a2, vece);
3287        break;
3288    case INDEX_op_umin_vec:
3289        tcg_out_insn(s, VRRc, VMNL, a0, a1, a2, vece);
3290        break;
3291    case INDEX_op_umax_vec:
3292        tcg_out_insn(s, VRRc, VMXL, a0, a1, a2, vece);
3293        break;
3294
3295    case INDEX_op_bitsel_vec:
3296        tcg_out_insn(s, VRRe, VSEL, a0, a2, args[3], a1);
3297        break;
3298
3299    case INDEX_op_cmp_vec:
3300        tcg_out_cmp_vec(s, vece, a0, a1, a2, args[3]);
3301        break;
3302    case INDEX_op_cmpsel_vec:
3303        tcg_out_cmpsel_vec(s, vece, a0, a1, a2, args[3], const_args[3],
3304                           args[4], args[5]);
3305        break;
3306
3307    case INDEX_op_s390_vuph_vec:
3308        tcg_out_insn(s, VRRa, VUPH, a0, a1, vece);
3309        break;
3310    case INDEX_op_s390_vupl_vec:
3311        tcg_out_insn(s, VRRa, VUPL, a0, a1, vece);
3312        break;
3313    case INDEX_op_s390_vpks_vec:
3314        tcg_out_insn(s, VRRc, VPKS, a0, a1, a2, vece);
3315        break;
3316
3317    case INDEX_op_mov_vec:   /* Always emitted via tcg_out_mov.  */
3318    case INDEX_op_dup_vec:   /* Always emitted via tcg_out_dup_vec.  */
3319    default:
3320        g_assert_not_reached();
3321    }
3322}
3323
3324int tcg_can_emit_vec_op(TCGOpcode opc, TCGType type, unsigned vece)
3325{
3326    switch (opc) {
3327    case INDEX_op_abs_vec:
3328    case INDEX_op_add_vec:
3329    case INDEX_op_and_vec:
3330    case INDEX_op_andc_vec:
3331    case INDEX_op_bitsel_vec:
3332    case INDEX_op_eqv_vec:
3333    case INDEX_op_nand_vec:
3334    case INDEX_op_neg_vec:
3335    case INDEX_op_nor_vec:
3336    case INDEX_op_not_vec:
3337    case INDEX_op_or_vec:
3338    case INDEX_op_orc_vec:
3339    case INDEX_op_rotli_vec:
3340    case INDEX_op_rotls_vec:
3341    case INDEX_op_rotlv_vec:
3342    case INDEX_op_sari_vec:
3343    case INDEX_op_sars_vec:
3344    case INDEX_op_sarv_vec:
3345    case INDEX_op_shli_vec:
3346    case INDEX_op_shls_vec:
3347    case INDEX_op_shlv_vec:
3348    case INDEX_op_shri_vec:
3349    case INDEX_op_shrs_vec:
3350    case INDEX_op_shrv_vec:
3351    case INDEX_op_smax_vec:
3352    case INDEX_op_smin_vec:
3353    case INDEX_op_sub_vec:
3354    case INDEX_op_umax_vec:
3355    case INDEX_op_umin_vec:
3356    case INDEX_op_xor_vec:
3357    case INDEX_op_cmp_vec:
3358    case INDEX_op_cmpsel_vec:
3359        return 1;
3360    case INDEX_op_rotrv_vec:
3361        return -1;
3362    case INDEX_op_mul_vec:
3363        return vece < MO_64;
3364    case INDEX_op_ssadd_vec:
3365    case INDEX_op_sssub_vec:
3366        return vece < MO_64 ? -1 : 0;
3367    default:
3368        return 0;
3369    }
3370}
3371
3372static void expand_vec_sat(TCGType type, unsigned vece, TCGv_vec v0,
3373                           TCGv_vec v1, TCGv_vec v2, TCGOpcode add_sub_opc)
3374{
3375    TCGv_vec h1 = tcg_temp_new_vec(type);
3376    TCGv_vec h2 = tcg_temp_new_vec(type);
3377    TCGv_vec l1 = tcg_temp_new_vec(type);
3378    TCGv_vec l2 = tcg_temp_new_vec(type);
3379
3380    tcg_debug_assert (vece < MO_64);
3381
3382    /* Unpack with sign-extension. */
3383    vec_gen_2(INDEX_op_s390_vuph_vec, type, vece,
3384              tcgv_vec_arg(h1), tcgv_vec_arg(v1));
3385    vec_gen_2(INDEX_op_s390_vuph_vec, type, vece,
3386              tcgv_vec_arg(h2), tcgv_vec_arg(v2));
3387
3388    vec_gen_2(INDEX_op_s390_vupl_vec, type, vece,
3389              tcgv_vec_arg(l1), tcgv_vec_arg(v1));
3390    vec_gen_2(INDEX_op_s390_vupl_vec, type, vece,
3391              tcgv_vec_arg(l2), tcgv_vec_arg(v2));
3392
3393    /* Arithmetic on a wider element size. */
3394    vec_gen_3(add_sub_opc, type, vece + 1, tcgv_vec_arg(h1),
3395              tcgv_vec_arg(h1), tcgv_vec_arg(h2));
3396    vec_gen_3(add_sub_opc, type, vece + 1, tcgv_vec_arg(l1),
3397              tcgv_vec_arg(l1), tcgv_vec_arg(l2));
3398
3399    /* Pack with saturation. */
3400    vec_gen_3(INDEX_op_s390_vpks_vec, type, vece + 1,
3401              tcgv_vec_arg(v0), tcgv_vec_arg(h1), tcgv_vec_arg(l1));
3402
3403    tcg_temp_free_vec(h1);
3404    tcg_temp_free_vec(h2);
3405    tcg_temp_free_vec(l1);
3406    tcg_temp_free_vec(l2);
3407}
3408
3409void tcg_expand_vec_op(TCGOpcode opc, TCGType type, unsigned vece,
3410                       TCGArg a0, ...)
3411{
3412    va_list va;
3413    TCGv_vec v0, v1, v2, t0;
3414
3415    va_start(va, a0);
3416    v0 = temp_tcgv_vec(arg_temp(a0));
3417    v1 = temp_tcgv_vec(arg_temp(va_arg(va, TCGArg)));
3418    v2 = temp_tcgv_vec(arg_temp(va_arg(va, TCGArg)));
3419
3420    switch (opc) {
3421    case INDEX_op_rotrv_vec:
3422        t0 = tcg_temp_new_vec(type);
3423        tcg_gen_neg_vec(vece, t0, v2);
3424        tcg_gen_rotlv_vec(vece, v0, v1, t0);
3425        tcg_temp_free_vec(t0);
3426        break;
3427
3428    case INDEX_op_ssadd_vec:
3429        expand_vec_sat(type, vece, v0, v1, v2, INDEX_op_add_vec);
3430        break;
3431    case INDEX_op_sssub_vec:
3432        expand_vec_sat(type, vece, v0, v1, v2, INDEX_op_sub_vec);
3433        break;
3434
3435    default:
3436        g_assert_not_reached();
3437    }
3438    va_end(va);
3439}
3440
3441static TCGConstraintSetIndex
3442tcg_target_op_def(TCGOpcode op, TCGType type, unsigned flags)
3443{
3444    switch (op) {
3445    case INDEX_op_goto_ptr:
3446        return C_O0_I1(r);
3447
3448    case INDEX_op_ld8u_i32:
3449    case INDEX_op_ld8u_i64:
3450    case INDEX_op_ld8s_i32:
3451    case INDEX_op_ld8s_i64:
3452    case INDEX_op_ld16u_i32:
3453    case INDEX_op_ld16u_i64:
3454    case INDEX_op_ld16s_i32:
3455    case INDEX_op_ld16s_i64:
3456    case INDEX_op_ld_i32:
3457    case INDEX_op_ld32u_i64:
3458    case INDEX_op_ld32s_i64:
3459    case INDEX_op_ld_i64:
3460        return C_O1_I1(r, r);
3461
3462    case INDEX_op_st8_i32:
3463    case INDEX_op_st8_i64:
3464    case INDEX_op_st16_i32:
3465    case INDEX_op_st16_i64:
3466    case INDEX_op_st_i32:
3467    case INDEX_op_st32_i64:
3468    case INDEX_op_st_i64:
3469        return C_O0_I2(r, r);
3470
3471    case INDEX_op_ext_i32_i64:
3472    case INDEX_op_extu_i32_i64:
3473    case INDEX_op_extract_i32:
3474    case INDEX_op_extract_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