xref: /openbmc/qemu/tcg/riscv/tcg-target.c.inc (revision 63070ce368e1a2d430b9022a9db46f1817628efc)
1/*
2 * Tiny Code Generator for QEMU
3 *
4 * Copyright (c) 2018 SiFive, Inc
5 * Copyright (c) 2008-2009 Arnaud Patard <arnaud.patard@rtp-net.org>
6 * Copyright (c) 2009 Aurelien Jarno <aurelien@aurel32.net>
7 * Copyright (c) 2008 Fabrice Bellard
8 *
9 * Based on i386/tcg-target.c and mips/tcg-target.c
10 *
11 * Permission is hereby granted, free of charge, to any person obtaining a copy
12 * of this software and associated documentation files (the "Software"), to deal
13 * in the Software without restriction, including without limitation the rights
14 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
15 * copies of the Software, and to permit persons to whom the Software is
16 * furnished to do so, subject to the following conditions:
17 *
18 * The above copyright notice and this permission notice shall be included in
19 * all copies or substantial portions of the Software.
20 *
21 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
22 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
23 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
24 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
25 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
26 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
27 * THE SOFTWARE.
28 */
29
30/* Used for function call generation. */
31#define TCG_REG_CALL_STACK              TCG_REG_SP
32#define TCG_TARGET_STACK_ALIGN          16
33#define TCG_TARGET_CALL_STACK_OFFSET    0
34#define TCG_TARGET_CALL_ARG_I32         TCG_CALL_ARG_NORMAL
35#define TCG_TARGET_CALL_ARG_I64         TCG_CALL_ARG_NORMAL
36#define TCG_TARGET_CALL_ARG_I128        TCG_CALL_ARG_NORMAL
37#define TCG_TARGET_CALL_RET_I128        TCG_CALL_RET_NORMAL
38
39#ifdef CONFIG_DEBUG_TCG
40static const char * const tcg_target_reg_names[TCG_TARGET_NB_REGS] = {
41    "zero", "ra",  "sp",  "gp",  "tp",  "t0",  "t1",  "t2",
42    "s0",   "s1",  "a0",  "a1",  "a2",  "a3",  "a4",  "a5",
43    "a6",   "a7",  "s2",  "s3",  "s4",  "s5",  "s6",  "s7",
44    "s8",   "s9",  "s10", "s11", "t3",  "t4",  "t5",  "t6",
45    "v0",   "v1",  "v2",  "v3",  "v4",  "v5",  "v6",  "v7",
46    "v8",   "v9",  "v10", "v11", "v12", "v13", "v14", "v15",
47    "v16",  "v17", "v18", "v19", "v20", "v21", "v22", "v23",
48    "v24",  "v25", "v26", "v27", "v28", "v29", "v30", "v31",
49};
50#endif
51
52static const int tcg_target_reg_alloc_order[] = {
53    /* Call saved registers */
54    /* TCG_REG_S0 reserved for TCG_AREG0 */
55    TCG_REG_S1,
56    TCG_REG_S2,
57    TCG_REG_S3,
58    TCG_REG_S4,
59    TCG_REG_S5,
60    TCG_REG_S6,
61    TCG_REG_S7,
62    TCG_REG_S8,
63    TCG_REG_S9,
64    TCG_REG_S10,
65    TCG_REG_S11,
66
67    /* Call clobbered registers */
68    TCG_REG_T0,
69    TCG_REG_T1,
70    TCG_REG_T2,
71    TCG_REG_T3,
72    TCG_REG_T4,
73    TCG_REG_T5,
74    TCG_REG_T6,
75
76    /* Argument registers */
77    TCG_REG_A0,
78    TCG_REG_A1,
79    TCG_REG_A2,
80    TCG_REG_A3,
81    TCG_REG_A4,
82    TCG_REG_A5,
83    TCG_REG_A6,
84    TCG_REG_A7,
85
86    /* Vector registers and TCG_REG_V0 reserved for mask. */
87    TCG_REG_V1,  TCG_REG_V2,  TCG_REG_V3,  TCG_REG_V4,
88    TCG_REG_V5,  TCG_REG_V6,  TCG_REG_V7,  TCG_REG_V8,
89    TCG_REG_V9,  TCG_REG_V10, TCG_REG_V11, TCG_REG_V12,
90    TCG_REG_V13, TCG_REG_V14, TCG_REG_V15, TCG_REG_V16,
91    TCG_REG_V17, TCG_REG_V18, TCG_REG_V19, TCG_REG_V20,
92    TCG_REG_V21, TCG_REG_V22, TCG_REG_V23, TCG_REG_V24,
93    TCG_REG_V25, TCG_REG_V26, TCG_REG_V27, TCG_REG_V28,
94    TCG_REG_V29, TCG_REG_V30, TCG_REG_V31,
95};
96
97static const int tcg_target_call_iarg_regs[] = {
98    TCG_REG_A0,
99    TCG_REG_A1,
100    TCG_REG_A2,
101    TCG_REG_A3,
102    TCG_REG_A4,
103    TCG_REG_A5,
104    TCG_REG_A6,
105    TCG_REG_A7,
106};
107
108static TCGReg tcg_target_call_oarg_reg(TCGCallReturnKind kind, int slot)
109{
110    tcg_debug_assert(kind == TCG_CALL_RET_NORMAL);
111    tcg_debug_assert(slot >= 0 && slot <= 1);
112    return TCG_REG_A0 + slot;
113}
114
115#define TCG_CT_CONST_S12     0x100
116#define TCG_CT_CONST_M12     0x200
117#define TCG_CT_CONST_S5      0x400
118#define TCG_CT_CONST_CMP_VI  0x800
119
120#define ALL_GENERAL_REGS   MAKE_64BIT_MASK(0, 32)
121#define ALL_VECTOR_REGS    MAKE_64BIT_MASK(32, 32)
122#define ALL_DVECTOR_REG_GROUPS 0x5555555500000000
123#define ALL_QVECTOR_REG_GROUPS 0x1111111100000000
124
125#define sextreg  sextract64
126
127/*
128 * RISC-V Base ISA opcodes (IM)
129 */
130
131#define V_OPIVV (0x0 << 12)
132#define V_OPFVV (0x1 << 12)
133#define V_OPMVV (0x2 << 12)
134#define V_OPIVI (0x3 << 12)
135#define V_OPIVX (0x4 << 12)
136#define V_OPFVF (0x5 << 12)
137#define V_OPMVX (0x6 << 12)
138#define V_OPCFG (0x7 << 12)
139
140/* NF <= 7 && NF >= 0 */
141#define V_NF(x) (x << 29)
142#define V_UNIT_STRIDE (0x0 << 20)
143#define V_UNIT_STRIDE_WHOLE_REG (0x8 << 20)
144
145typedef enum {
146    VLMUL_M1 = 0, /* LMUL=1 */
147    VLMUL_M2,     /* LMUL=2 */
148    VLMUL_M4,     /* LMUL=4 */
149    VLMUL_M8,     /* LMUL=8 */
150    VLMUL_RESERVED,
151    VLMUL_MF8,    /* LMUL=1/8 */
152    VLMUL_MF4,    /* LMUL=1/4 */
153    VLMUL_MF2,    /* LMUL=1/2 */
154} RISCVVlmul;
155
156typedef enum {
157    OPC_ADD = 0x33,
158    OPC_ADDI = 0x13,
159    OPC_AND = 0x7033,
160    OPC_ANDI = 0x7013,
161    OPC_AUIPC = 0x17,
162    OPC_BEQ = 0x63,
163    OPC_BEXTI = 0x48005013,
164    OPC_BGE = 0x5063,
165    OPC_BGEU = 0x7063,
166    OPC_BLT = 0x4063,
167    OPC_BLTU = 0x6063,
168    OPC_BNE = 0x1063,
169    OPC_DIV = 0x2004033,
170    OPC_DIVU = 0x2005033,
171    OPC_JAL = 0x6f,
172    OPC_JALR = 0x67,
173    OPC_LB = 0x3,
174    OPC_LBU = 0x4003,
175    OPC_LD = 0x3003,
176    OPC_LH = 0x1003,
177    OPC_LHU = 0x5003,
178    OPC_LUI = 0x37,
179    OPC_LW = 0x2003,
180    OPC_LWU = 0x6003,
181    OPC_MUL = 0x2000033,
182    OPC_MULH = 0x2001033,
183    OPC_MULHSU = 0x2002033,
184    OPC_MULHU = 0x2003033,
185    OPC_OR = 0x6033,
186    OPC_ORI = 0x6013,
187    OPC_REM = 0x2006033,
188    OPC_REMU = 0x2007033,
189    OPC_SB = 0x23,
190    OPC_SD = 0x3023,
191    OPC_SH = 0x1023,
192    OPC_SLL = 0x1033,
193    OPC_SLLI = 0x1013,
194    OPC_SLT = 0x2033,
195    OPC_SLTI = 0x2013,
196    OPC_SLTIU = 0x3013,
197    OPC_SLTU = 0x3033,
198    OPC_SRA = 0x40005033,
199    OPC_SRAI = 0x40005013,
200    OPC_SRL = 0x5033,
201    OPC_SRLI = 0x5013,
202    OPC_SUB = 0x40000033,
203    OPC_SW = 0x2023,
204    OPC_XOR = 0x4033,
205    OPC_XORI = 0x4013,
206
207    OPC_ADDIW = 0x1b,
208    OPC_ADDW = 0x3b,
209    OPC_DIVUW = 0x200503b,
210    OPC_DIVW = 0x200403b,
211    OPC_MULW = 0x200003b,
212    OPC_REMUW = 0x200703b,
213    OPC_REMW = 0x200603b,
214    OPC_SLLIW = 0x101b,
215    OPC_SLLW = 0x103b,
216    OPC_SRAIW = 0x4000501b,
217    OPC_SRAW = 0x4000503b,
218    OPC_SRLIW = 0x501b,
219    OPC_SRLW = 0x503b,
220    OPC_SUBW = 0x4000003b,
221
222    OPC_FENCE = 0x0000000f,
223    OPC_NOP   = OPC_ADDI,   /* nop = addi r0,r0,0 */
224
225    /* Zba: Bit manipulation extension, address generation */
226    OPC_ADD_UW = 0x0800003b,
227
228    /* Zbb: Bit manipulation extension, basic bit manipulation */
229    OPC_ANDN   = 0x40007033,
230    OPC_CLZ    = 0x60001013,
231    OPC_CLZW   = 0x6000101b,
232    OPC_CPOP   = 0x60201013,
233    OPC_CPOPW  = 0x6020101b,
234    OPC_CTZ    = 0x60101013,
235    OPC_CTZW   = 0x6010101b,
236    OPC_ORN    = 0x40006033,
237    OPC_REV8   = 0x6b805013,
238    OPC_ROL    = 0x60001033,
239    OPC_ROLW   = 0x6000103b,
240    OPC_ROR    = 0x60005033,
241    OPC_RORW   = 0x6000503b,
242    OPC_RORI   = 0x60005013,
243    OPC_RORIW  = 0x6000501b,
244    OPC_SEXT_B = 0x60401013,
245    OPC_SEXT_H = 0x60501013,
246    OPC_XNOR   = 0x40004033,
247    OPC_ZEXT_H = 0x0800403b,
248
249    /* Zicond: integer conditional operations */
250    OPC_CZERO_EQZ = 0x0e005033,
251    OPC_CZERO_NEZ = 0x0e007033,
252
253    /* V: Vector extension 1.0 */
254    OPC_VSETVLI  = 0x57 | V_OPCFG,
255    OPC_VSETIVLI = 0xc0000057 | V_OPCFG,
256    OPC_VSETVL   = 0x80000057 | V_OPCFG,
257
258    OPC_VLE8_V  = 0x7 | V_UNIT_STRIDE,
259    OPC_VLE16_V = 0x5007 | V_UNIT_STRIDE,
260    OPC_VLE32_V = 0x6007 | V_UNIT_STRIDE,
261    OPC_VLE64_V = 0x7007 | V_UNIT_STRIDE,
262    OPC_VSE8_V  = 0x27 | V_UNIT_STRIDE,
263    OPC_VSE16_V = 0x5027 | V_UNIT_STRIDE,
264    OPC_VSE32_V = 0x6027 | V_UNIT_STRIDE,
265    OPC_VSE64_V = 0x7027 | V_UNIT_STRIDE,
266
267    OPC_VL1RE64_V = 0x2007007 | V_UNIT_STRIDE_WHOLE_REG | V_NF(0),
268    OPC_VL2RE64_V = 0x2007007 | V_UNIT_STRIDE_WHOLE_REG | V_NF(1),
269    OPC_VL4RE64_V = 0x2007007 | V_UNIT_STRIDE_WHOLE_REG | V_NF(3),
270    OPC_VL8RE64_V = 0x2007007 | V_UNIT_STRIDE_WHOLE_REG | V_NF(7),
271
272    OPC_VS1R_V = 0x2000027 | V_UNIT_STRIDE_WHOLE_REG | V_NF(0),
273    OPC_VS2R_V = 0x2000027 | V_UNIT_STRIDE_WHOLE_REG | V_NF(1),
274    OPC_VS4R_V = 0x2000027 | V_UNIT_STRIDE_WHOLE_REG | V_NF(3),
275    OPC_VS8R_V = 0x2000027 | V_UNIT_STRIDE_WHOLE_REG | V_NF(7),
276
277    OPC_VMERGE_VIM = 0x5c000057 | V_OPIVI,
278    OPC_VMERGE_VVM = 0x5c000057 | V_OPIVV,
279
280    OPC_VADD_VV = 0x57 | V_OPIVV,
281    OPC_VADD_VI = 0x57 | V_OPIVI,
282    OPC_VSUB_VV = 0x8000057 | V_OPIVV,
283    OPC_VRSUB_VI = 0xc000057 | V_OPIVI,
284    OPC_VAND_VV = 0x24000057 | V_OPIVV,
285    OPC_VAND_VI = 0x24000057 | V_OPIVI,
286    OPC_VOR_VV = 0x28000057 | V_OPIVV,
287    OPC_VOR_VI = 0x28000057 | V_OPIVI,
288    OPC_VXOR_VV = 0x2c000057 | V_OPIVV,
289    OPC_VXOR_VI = 0x2c000057 | V_OPIVI,
290
291    OPC_VMUL_VV = 0x94000057 | V_OPMVV,
292    OPC_VSADD_VV = 0x84000057 | V_OPIVV,
293    OPC_VSADD_VI = 0x84000057 | V_OPIVI,
294    OPC_VSSUB_VV = 0x8c000057 | V_OPIVV,
295    OPC_VSSUB_VI = 0x8c000057 | V_OPIVI,
296    OPC_VSADDU_VV = 0x80000057 | V_OPIVV,
297    OPC_VSADDU_VI = 0x80000057 | V_OPIVI,
298    OPC_VSSUBU_VV = 0x88000057 | V_OPIVV,
299    OPC_VSSUBU_VI = 0x88000057 | V_OPIVI,
300
301    OPC_VMAX_VV = 0x1c000057 | V_OPIVV,
302    OPC_VMAX_VI = 0x1c000057 | V_OPIVI,
303    OPC_VMAXU_VV = 0x18000057 | V_OPIVV,
304    OPC_VMAXU_VI = 0x18000057 | V_OPIVI,
305    OPC_VMIN_VV = 0x14000057 | V_OPIVV,
306    OPC_VMIN_VI = 0x14000057 | V_OPIVI,
307    OPC_VMINU_VV = 0x10000057 | V_OPIVV,
308    OPC_VMINU_VI = 0x10000057 | V_OPIVI,
309
310    OPC_VMSEQ_VV = 0x60000057 | V_OPIVV,
311    OPC_VMSEQ_VI = 0x60000057 | V_OPIVI,
312    OPC_VMSEQ_VX = 0x60000057 | V_OPIVX,
313    OPC_VMSNE_VV = 0x64000057 | V_OPIVV,
314    OPC_VMSNE_VI = 0x64000057 | V_OPIVI,
315    OPC_VMSNE_VX = 0x64000057 | V_OPIVX,
316
317    OPC_VMSLTU_VV = 0x68000057 | V_OPIVV,
318    OPC_VMSLTU_VX = 0x68000057 | V_OPIVX,
319    OPC_VMSLT_VV = 0x6c000057 | V_OPIVV,
320    OPC_VMSLT_VX = 0x6c000057 | V_OPIVX,
321    OPC_VMSLEU_VV = 0x70000057 | V_OPIVV,
322    OPC_VMSLEU_VX = 0x70000057 | V_OPIVX,
323    OPC_VMSLE_VV = 0x74000057 | V_OPIVV,
324    OPC_VMSLE_VX = 0x74000057 | V_OPIVX,
325
326    OPC_VMSLEU_VI = 0x70000057 | V_OPIVI,
327    OPC_VMSLE_VI = 0x74000057 | V_OPIVI,
328    OPC_VMSGTU_VI = 0x78000057 | V_OPIVI,
329    OPC_VMSGTU_VX = 0x78000057 | V_OPIVX,
330    OPC_VMSGT_VI = 0x7c000057 | V_OPIVI,
331    OPC_VMSGT_VX = 0x7c000057 | V_OPIVX,
332
333    OPC_VSLL_VV = 0x94000057 | V_OPIVV,
334    OPC_VSLL_VI = 0x94000057 | V_OPIVI,
335    OPC_VSLL_VX = 0x94000057 | V_OPIVX,
336    OPC_VSRL_VV = 0xa0000057 | V_OPIVV,
337    OPC_VSRL_VI = 0xa0000057 | V_OPIVI,
338    OPC_VSRL_VX = 0xa0000057 | V_OPIVX,
339    OPC_VSRA_VV = 0xa4000057 | V_OPIVV,
340    OPC_VSRA_VI = 0xa4000057 | V_OPIVI,
341    OPC_VSRA_VX = 0xa4000057 | V_OPIVX,
342
343    OPC_VMV_V_V = 0x5e000057 | V_OPIVV,
344    OPC_VMV_V_I = 0x5e000057 | V_OPIVI,
345    OPC_VMV_V_X = 0x5e000057 | V_OPIVX,
346
347    OPC_VMVNR_V = 0x9e000057 | V_OPIVI,
348} RISCVInsn;
349
350static const struct {
351    RISCVInsn op;
352    bool swap;
353} tcg_cmpcond_to_rvv_vv[] = {
354    [TCG_COND_EQ] =  { OPC_VMSEQ_VV,  false },
355    [TCG_COND_NE] =  { OPC_VMSNE_VV,  false },
356    [TCG_COND_LT] =  { OPC_VMSLT_VV,  false },
357    [TCG_COND_GE] =  { OPC_VMSLE_VV,  true  },
358    [TCG_COND_GT] =  { OPC_VMSLT_VV,  true  },
359    [TCG_COND_LE] =  { OPC_VMSLE_VV,  false },
360    [TCG_COND_LTU] = { OPC_VMSLTU_VV, false },
361    [TCG_COND_GEU] = { OPC_VMSLEU_VV, true  },
362    [TCG_COND_GTU] = { OPC_VMSLTU_VV, true  },
363    [TCG_COND_LEU] = { OPC_VMSLEU_VV, false }
364};
365
366static const struct {
367    RISCVInsn op;
368    int min;
369    int max;
370    bool adjust;
371}  tcg_cmpcond_to_rvv_vi[] = {
372    [TCG_COND_EQ]  = { OPC_VMSEQ_VI,  -16, 15, false },
373    [TCG_COND_NE]  = { OPC_VMSNE_VI,  -16, 15, false },
374    [TCG_COND_GT]  = { OPC_VMSGT_VI,  -16, 15, false },
375    [TCG_COND_LE]  = { OPC_VMSLE_VI,  -16, 15, false },
376    [TCG_COND_LT]  = { OPC_VMSLE_VI,  -15, 16, true  },
377    [TCG_COND_GE]  = { OPC_VMSGT_VI,  -15, 16, true  },
378    [TCG_COND_LEU] = { OPC_VMSLEU_VI,   0, 15, false },
379    [TCG_COND_GTU] = { OPC_VMSGTU_VI,   0, 15, false },
380    [TCG_COND_LTU] = { OPC_VMSLEU_VI,   1, 16, true  },
381    [TCG_COND_GEU] = { OPC_VMSGTU_VI,   1, 16, true  },
382};
383
384/* test if a constant matches the constraint */
385static bool tcg_target_const_match(int64_t val, int ct,
386                                   TCGType type, TCGCond cond, int vece)
387{
388    if (ct & TCG_CT_CONST) {
389        return 1;
390    }
391    if (type >= TCG_TYPE_V64) {
392        /* Val is replicated by VECE; extract the highest element. */
393        val >>= (-8 << vece) & 63;
394    }
395    /*
396     * Sign extended from 12 bits: [-0x800, 0x7ff].
397     * Used for most arithmetic, as this is the isa field.
398     */
399    if ((ct & TCG_CT_CONST_S12) && val >= -0x800 && val <= 0x7ff) {
400        return 1;
401    }
402    /*
403     * Sign extended from 12 bits, +/- matching: [-0x7ff, 0x7ff].
404     * Used by movcond, which may need the negative value,
405     * and requires the modified constant to be representable.
406     */
407    if ((ct & TCG_CT_CONST_M12) && val >= -0x7ff && val <= 0x7ff) {
408        return 1;
409    }
410    /*
411     * Sign extended from 5 bits: [-0x10, 0x0f].
412     * Used for vector-immediate.
413     */
414    if ((ct & TCG_CT_CONST_S5) && val >= -0x10 && val <= 0x0f) {
415        return 1;
416    }
417    /*
418     * Used for vector compare OPIVI instructions.
419     */
420    if ((ct & TCG_CT_CONST_CMP_VI) &&
421        val >= tcg_cmpcond_to_rvv_vi[cond].min &&
422        val <= tcg_cmpcond_to_rvv_vi[cond].max) {
423        return true;
424     }
425    return 0;
426}
427
428/*
429 * RISC-V immediate and instruction encoders (excludes 16-bit RVC)
430 */
431
432/* Type-R */
433
434static int32_t encode_r(RISCVInsn opc, TCGReg rd, TCGReg rs1, TCGReg rs2)
435{
436    return opc | (rd & 0x1f) << 7 | (rs1 & 0x1f) << 15 | (rs2 & 0x1f) << 20;
437}
438
439/* Type-I */
440
441static int32_t encode_imm12(uint32_t imm)
442{
443    return (imm & 0xfff) << 20;
444}
445
446static int32_t encode_i(RISCVInsn opc, TCGReg rd, TCGReg rs1, uint32_t imm)
447{
448    return opc | (rd & 0x1f) << 7 | (rs1 & 0x1f) << 15 | encode_imm12(imm);
449}
450
451/* Type-S */
452
453static int32_t encode_simm12(uint32_t imm)
454{
455    int32_t ret = 0;
456
457    ret |= (imm & 0xFE0) << 20;
458    ret |= (imm & 0x1F) << 7;
459
460    return ret;
461}
462
463static int32_t encode_s(RISCVInsn opc, TCGReg rs1, TCGReg rs2, uint32_t imm)
464{
465    return opc | (rs1 & 0x1f) << 15 | (rs2 & 0x1f) << 20 | encode_simm12(imm);
466}
467
468/* Type-SB */
469
470static int32_t encode_sbimm12(uint32_t imm)
471{
472    int32_t ret = 0;
473
474    ret |= (imm & 0x1000) << 19;
475    ret |= (imm & 0x7e0) << 20;
476    ret |= (imm & 0x1e) << 7;
477    ret |= (imm & 0x800) >> 4;
478
479    return ret;
480}
481
482static int32_t encode_sb(RISCVInsn opc, TCGReg rs1, TCGReg rs2, uint32_t imm)
483{
484    return opc | (rs1 & 0x1f) << 15 | (rs2 & 0x1f) << 20 | encode_sbimm12(imm);
485}
486
487/* Type-U */
488
489static int32_t encode_uimm20(uint32_t imm)
490{
491    return imm & 0xfffff000;
492}
493
494static int32_t encode_u(RISCVInsn opc, TCGReg rd, uint32_t imm)
495{
496    return opc | (rd & 0x1f) << 7 | encode_uimm20(imm);
497}
498
499/* Type-UJ */
500
501static int32_t encode_ujimm20(uint32_t imm)
502{
503    int32_t ret = 0;
504
505    ret |= (imm & 0x0007fe) << (21 - 1);
506    ret |= (imm & 0x000800) << (20 - 11);
507    ret |= (imm & 0x0ff000) << (12 - 12);
508    ret |= (imm & 0x100000) << (31 - 20);
509
510    return ret;
511}
512
513static int32_t encode_uj(RISCVInsn opc, TCGReg rd, uint32_t imm)
514{
515    return opc | (rd & 0x1f) << 7 | encode_ujimm20(imm);
516}
517
518
519/* Type-OPIVI */
520
521static int32_t encode_vi(RISCVInsn opc, TCGReg rd, int32_t imm,
522                         TCGReg vs2, bool vm)
523{
524    return opc | (rd & 0x1f) << 7 | (imm & 0x1f) << 15 |
525           (vs2 & 0x1f) << 20 | (vm << 25);
526}
527
528/* Type-OPIVV/OPMVV/OPIVX/OPMVX, Vector load and store */
529
530static int32_t encode_v(RISCVInsn opc, TCGReg d, TCGReg s1,
531                        TCGReg s2, bool vm)
532{
533    return opc | (d & 0x1f) << 7 | (s1 & 0x1f) << 15 |
534           (s2 & 0x1f) << 20 | (vm << 25);
535}
536
537/* Vector vtype */
538
539static uint32_t encode_vtype(bool vta, bool vma,
540                            MemOp vsew, RISCVVlmul vlmul)
541{
542    return vma << 7 | vta << 6 | vsew << 3 | vlmul;
543}
544
545static int32_t encode_vset(RISCVInsn opc, TCGReg rd,
546                           TCGArg rs1, uint32_t vtype)
547{
548    return opc | (rd & 0x1f) << 7 | (rs1 & 0x1f) << 15 | (vtype & 0x7ff) << 20;
549}
550
551static int32_t encode_vseti(RISCVInsn opc, TCGReg rd,
552                            uint32_t uimm, uint32_t vtype)
553{
554    return opc | (rd & 0x1f) << 7 | (uimm & 0x1f) << 15 | (vtype & 0x3ff) << 20;
555}
556
557/*
558 * RISC-V instruction emitters
559 */
560
561static void tcg_out_opc_reg(TCGContext *s, RISCVInsn opc,
562                            TCGReg rd, TCGReg rs1, TCGReg rs2)
563{
564    tcg_out32(s, encode_r(opc, rd, rs1, rs2));
565}
566
567static void tcg_out_opc_imm(TCGContext *s, RISCVInsn opc,
568                            TCGReg rd, TCGReg rs1, TCGArg imm)
569{
570    tcg_out32(s, encode_i(opc, rd, rs1, imm));
571}
572
573static void tcg_out_opc_store(TCGContext *s, RISCVInsn opc,
574                              TCGReg rs1, TCGReg rs2, uint32_t imm)
575{
576    tcg_out32(s, encode_s(opc, rs1, rs2, imm));
577}
578
579static void tcg_out_opc_branch(TCGContext *s, RISCVInsn opc,
580                               TCGReg rs1, TCGReg rs2, uint32_t imm)
581{
582    tcg_out32(s, encode_sb(opc, rs1, rs2, imm));
583}
584
585static void tcg_out_opc_upper(TCGContext *s, RISCVInsn opc,
586                              TCGReg rd, uint32_t imm)
587{
588    tcg_out32(s, encode_u(opc, rd, imm));
589}
590
591static void tcg_out_opc_jump(TCGContext *s, RISCVInsn opc,
592                             TCGReg rd, uint32_t imm)
593{
594    tcg_out32(s, encode_uj(opc, rd, imm));
595}
596
597static void tcg_out_nop_fill(tcg_insn_unit *p, int count)
598{
599    int i;
600    for (i = 0; i < count; ++i) {
601        p[i] = OPC_NOP;
602    }
603}
604
605/*
606 * Relocations
607 */
608
609static bool reloc_sbimm12(tcg_insn_unit *src_rw, const tcg_insn_unit *target)
610{
611    const tcg_insn_unit *src_rx = tcg_splitwx_to_rx(src_rw);
612    intptr_t offset = (intptr_t)target - (intptr_t)src_rx;
613
614    tcg_debug_assert((offset & 1) == 0);
615    if (offset == sextreg(offset, 0, 12)) {
616        *src_rw |= encode_sbimm12(offset);
617        return true;
618    }
619
620    return false;
621}
622
623static bool reloc_jimm20(tcg_insn_unit *src_rw, const tcg_insn_unit *target)
624{
625    const tcg_insn_unit *src_rx = tcg_splitwx_to_rx(src_rw);
626    intptr_t offset = (intptr_t)target - (intptr_t)src_rx;
627
628    tcg_debug_assert((offset & 1) == 0);
629    if (offset == sextreg(offset, 0, 20)) {
630        *src_rw |= encode_ujimm20(offset);
631        return true;
632    }
633
634    return false;
635}
636
637static bool reloc_call(tcg_insn_unit *src_rw, const tcg_insn_unit *target)
638{
639    const tcg_insn_unit *src_rx = tcg_splitwx_to_rx(src_rw);
640    intptr_t offset = (intptr_t)target - (intptr_t)src_rx;
641    int32_t lo = sextreg(offset, 0, 12);
642    int32_t hi = offset - lo;
643
644    if (offset == hi + lo) {
645        src_rw[0] |= encode_uimm20(hi);
646        src_rw[1] |= encode_imm12(lo);
647        return true;
648    }
649
650    return false;
651}
652
653static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
654                        intptr_t value, intptr_t addend)
655{
656    tcg_debug_assert(addend == 0);
657    switch (type) {
658    case R_RISCV_BRANCH:
659        return reloc_sbimm12(code_ptr, (tcg_insn_unit *)value);
660    case R_RISCV_JAL:
661        return reloc_jimm20(code_ptr, (tcg_insn_unit *)value);
662    case R_RISCV_CALL:
663        return reloc_call(code_ptr, (tcg_insn_unit *)value);
664    default:
665        g_assert_not_reached();
666    }
667}
668
669/*
670 * RISC-V vector instruction emitters
671 */
672
673/*
674 * Vector registers uses the same 5 lower bits as GPR registers,
675 * and vm=0 (vm = false) means vector masking ENABLED.
676 * With RVV 1.0, vs2 is the first operand, while rs1/imm is the
677 * second operand.
678 */
679static void tcg_out_opc_vv(TCGContext *s, RISCVInsn opc,
680                           TCGReg vd, TCGReg vs2, TCGReg vs1)
681{
682    tcg_out32(s, encode_v(opc, vd, vs1, vs2, true));
683}
684
685static void tcg_out_opc_vx(TCGContext *s, RISCVInsn opc,
686                           TCGReg vd, TCGReg vs2, TCGReg rs1)
687{
688    tcg_out32(s, encode_v(opc, vd, rs1, vs2, true));
689}
690
691static void tcg_out_opc_vi(TCGContext *s, RISCVInsn opc,
692                           TCGReg vd, TCGReg vs2, int32_t imm)
693{
694    tcg_out32(s, encode_vi(opc, vd, imm, vs2, true));
695}
696
697static void tcg_out_opc_vv_vi(TCGContext *s, RISCVInsn o_vv, RISCVInsn o_vi,
698                              TCGReg vd, TCGReg vs2, TCGArg vi1, int c_vi1)
699{
700    if (c_vi1) {
701        tcg_out_opc_vi(s, o_vi, vd, vs2, vi1);
702    } else {
703        tcg_out_opc_vv(s, o_vv, vd, vs2, vi1);
704    }
705}
706
707static void tcg_out_opc_vim_mask(TCGContext *s, RISCVInsn opc, TCGReg vd,
708                                 TCGReg vs2, int32_t imm)
709{
710    tcg_out32(s, encode_vi(opc, vd, imm, vs2, false));
711}
712
713static void tcg_out_opc_vvm_mask(TCGContext *s, RISCVInsn opc, TCGReg vd,
714                                 TCGReg vs2, TCGReg vs1)
715{
716    tcg_out32(s, encode_v(opc, vd, vs1, vs2, false));
717}
718
719typedef struct VsetCache {
720    uint32_t movi_insn;
721    uint32_t vset_insn;
722} VsetCache;
723
724static VsetCache riscv_vset_cache[3][4];
725
726static void set_vtype(TCGContext *s, TCGType type, MemOp vsew)
727{
728    const VsetCache *p = &riscv_vset_cache[type - TCG_TYPE_V64][vsew];
729
730    s->riscv_cur_type = type;
731    s->riscv_cur_vsew = vsew;
732
733    if (p->movi_insn) {
734        tcg_out32(s, p->movi_insn);
735    }
736    tcg_out32(s, p->vset_insn);
737}
738
739static MemOp set_vtype_len(TCGContext *s, TCGType type)
740{
741    if (type != s->riscv_cur_type) {
742        set_vtype(s, type, MO_64);
743    }
744    return s->riscv_cur_vsew;
745}
746
747static void set_vtype_len_sew(TCGContext *s, TCGType type, MemOp vsew)
748{
749    if (type != s->riscv_cur_type || vsew != s->riscv_cur_vsew) {
750        set_vtype(s, type, vsew);
751    }
752}
753
754/*
755 * TCG intrinsics
756 */
757
758static bool tcg_out_mov(TCGContext *s, TCGType type, TCGReg ret, TCGReg arg)
759{
760    if (ret == arg) {
761        return true;
762    }
763    switch (type) {
764    case TCG_TYPE_I32:
765    case TCG_TYPE_I64:
766        tcg_out_opc_imm(s, OPC_ADDI, ret, arg, 0);
767        break;
768    case TCG_TYPE_V64:
769    case TCG_TYPE_V128:
770    case TCG_TYPE_V256:
771        {
772            int lmul = type - riscv_lg2_vlenb;
773            int nf = 1 << MAX(lmul, 0);
774            tcg_out_opc_vi(s, OPC_VMVNR_V, ret, arg, nf - 1);
775        }
776        break;
777    default:
778        g_assert_not_reached();
779    }
780    return true;
781}
782
783static void tcg_out_movi(TCGContext *s, TCGType type, TCGReg rd,
784                         tcg_target_long val)
785{
786    tcg_target_long lo, hi, tmp;
787    int shift, ret;
788
789    if (type == TCG_TYPE_I32) {
790        val = (int32_t)val;
791    }
792
793    lo = sextreg(val, 0, 12);
794    if (val == lo) {
795        tcg_out_opc_imm(s, OPC_ADDI, rd, TCG_REG_ZERO, lo);
796        return;
797    }
798
799    hi = val - lo;
800    if (val == (int32_t)val) {
801        tcg_out_opc_upper(s, OPC_LUI, rd, hi);
802        if (lo != 0) {
803            tcg_out_opc_imm(s, OPC_ADDIW, rd, rd, lo);
804        }
805        return;
806    }
807
808    tmp = tcg_pcrel_diff(s, (void *)val);
809    if (tmp == (int32_t)tmp) {
810        tcg_out_opc_upper(s, OPC_AUIPC, rd, 0);
811        tcg_out_opc_imm(s, OPC_ADDI, rd, rd, 0);
812        ret = reloc_call(s->code_ptr - 2, (const tcg_insn_unit *)val);
813        tcg_debug_assert(ret == true);
814        return;
815    }
816
817    /* Look for a single 20-bit section.  */
818    shift = ctz64(val);
819    tmp = val >> shift;
820    if (tmp == sextreg(tmp, 0, 20)) {
821        tcg_out_opc_upper(s, OPC_LUI, rd, tmp << 12);
822        if (shift > 12) {
823            tcg_out_opc_imm(s, OPC_SLLI, rd, rd, shift - 12);
824        } else {
825            tcg_out_opc_imm(s, OPC_SRAI, rd, rd, 12 - shift);
826        }
827        return;
828    }
829
830    /* Look for a few high zero bits, with lots of bits set in the middle.  */
831    shift = clz64(val);
832    tmp = val << shift;
833    if (tmp == sextreg(tmp, 12, 20) << 12) {
834        tcg_out_opc_upper(s, OPC_LUI, rd, tmp);
835        tcg_out_opc_imm(s, OPC_SRLI, rd, rd, shift);
836        return;
837    } else if (tmp == sextreg(tmp, 0, 12)) {
838        tcg_out_opc_imm(s, OPC_ADDI, rd, TCG_REG_ZERO, tmp);
839        tcg_out_opc_imm(s, OPC_SRLI, rd, rd, shift);
840        return;
841    }
842
843    /* Drop into the constant pool.  */
844    new_pool_label(s, val, R_RISCV_CALL, s->code_ptr, 0);
845    tcg_out_opc_upper(s, OPC_AUIPC, rd, 0);
846    tcg_out_opc_imm(s, OPC_LD, rd, rd, 0);
847}
848
849static bool tcg_out_xchg(TCGContext *s, TCGType type, TCGReg r1, TCGReg r2)
850{
851    return false;
852}
853
854static void tcg_out_addi_ptr(TCGContext *s, TCGReg rd, TCGReg rs,
855                             tcg_target_long imm)
856{
857    /* This function is only used for passing structs by reference. */
858    g_assert_not_reached();
859}
860
861static void tcg_out_ext8u(TCGContext *s, TCGReg ret, TCGReg arg)
862{
863    tcg_out_opc_imm(s, OPC_ANDI, ret, arg, 0xff);
864}
865
866static void tcg_out_ext16u(TCGContext *s, TCGReg ret, TCGReg arg)
867{
868    if (cpuinfo & CPUINFO_ZBB) {
869        tcg_out_opc_reg(s, OPC_ZEXT_H, ret, arg, TCG_REG_ZERO);
870    } else {
871        tcg_out_opc_imm(s, OPC_SLLIW, ret, arg, 16);
872        tcg_out_opc_imm(s, OPC_SRLIW, ret, ret, 16);
873    }
874}
875
876static void tcg_out_ext32u(TCGContext *s, TCGReg ret, TCGReg arg)
877{
878    if (cpuinfo & CPUINFO_ZBA) {
879        tcg_out_opc_reg(s, OPC_ADD_UW, ret, arg, TCG_REG_ZERO);
880    } else {
881        tcg_out_opc_imm(s, OPC_SLLI, ret, arg, 32);
882        tcg_out_opc_imm(s, OPC_SRLI, ret, ret, 32);
883    }
884}
885
886static void tcg_out_ext8s(TCGContext *s, TCGType type, TCGReg ret, TCGReg arg)
887{
888    if (cpuinfo & CPUINFO_ZBB) {
889        tcg_out_opc_imm(s, OPC_SEXT_B, ret, arg, 0);
890    } else {
891        tcg_out_opc_imm(s, OPC_SLLIW, ret, arg, 24);
892        tcg_out_opc_imm(s, OPC_SRAIW, ret, ret, 24);
893    }
894}
895
896static void tcg_out_ext16s(TCGContext *s, TCGType type, TCGReg ret, TCGReg arg)
897{
898    if (cpuinfo & CPUINFO_ZBB) {
899        tcg_out_opc_imm(s, OPC_SEXT_H, ret, arg, 0);
900    } else {
901        tcg_out_opc_imm(s, OPC_SLLIW, ret, arg, 16);
902        tcg_out_opc_imm(s, OPC_SRAIW, ret, ret, 16);
903    }
904}
905
906static void tcg_out_ext32s(TCGContext *s, TCGReg ret, TCGReg arg)
907{
908    tcg_out_opc_imm(s, OPC_ADDIW, ret, arg, 0);
909}
910
911static void tcg_out_exts_i32_i64(TCGContext *s, TCGReg ret, TCGReg arg)
912{
913    if (ret != arg) {
914        tcg_out_ext32s(s, ret, arg);
915    }
916}
917
918static void tcg_out_extu_i32_i64(TCGContext *s, TCGReg ret, TCGReg arg)
919{
920    tcg_out_ext32u(s, ret, arg);
921}
922
923static void tcg_out_extrl_i64_i32(TCGContext *s, TCGReg ret, TCGReg arg)
924{
925    tcg_out_ext32s(s, ret, arg);
926}
927
928static void tcg_out_ldst(TCGContext *s, RISCVInsn opc, TCGReg data,
929                         TCGReg addr, intptr_t offset)
930{
931    intptr_t imm12 = sextreg(offset, 0, 12);
932
933    if (offset != imm12) {
934        intptr_t diff = tcg_pcrel_diff(s, (void *)offset);
935
936        if (addr == TCG_REG_ZERO && diff == (int32_t)diff) {
937            imm12 = sextreg(diff, 0, 12);
938            tcg_out_opc_upper(s, OPC_AUIPC, TCG_REG_TMP2, diff - imm12);
939        } else {
940            tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_TMP2, offset - imm12);
941            if (addr != TCG_REG_ZERO) {
942                tcg_out_opc_reg(s, OPC_ADD, TCG_REG_TMP2, TCG_REG_TMP2, addr);
943            }
944        }
945        addr = TCG_REG_TMP2;
946    }
947
948    switch (opc) {
949    case OPC_SB:
950    case OPC_SH:
951    case OPC_SW:
952    case OPC_SD:
953        tcg_out_opc_store(s, opc, addr, data, imm12);
954        break;
955    case OPC_LB:
956    case OPC_LBU:
957    case OPC_LH:
958    case OPC_LHU:
959    case OPC_LW:
960    case OPC_LWU:
961    case OPC_LD:
962        tcg_out_opc_imm(s, opc, data, addr, imm12);
963        break;
964    default:
965        g_assert_not_reached();
966    }
967}
968
969static void tcg_out_vec_ldst(TCGContext *s, RISCVInsn opc, TCGReg data,
970                             TCGReg addr, intptr_t offset)
971{
972    tcg_debug_assert(data >= TCG_REG_V0);
973    tcg_debug_assert(addr < TCG_REG_V0);
974
975    if (offset) {
976        tcg_debug_assert(addr != TCG_REG_ZERO);
977        if (offset == sextreg(offset, 0, 12)) {
978            tcg_out_opc_imm(s, OPC_ADDI, TCG_REG_TMP0, addr, offset);
979        } else {
980            tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_TMP0, offset);
981            tcg_out_opc_reg(s, OPC_ADD, TCG_REG_TMP0, TCG_REG_TMP0, addr);
982        }
983        addr = TCG_REG_TMP0;
984    }
985    tcg_out32(s, encode_v(opc, data, addr, 0, true));
986}
987
988static void tcg_out_ld(TCGContext *s, TCGType type, TCGReg arg,
989                       TCGReg arg1, intptr_t arg2)
990{
991    RISCVInsn insn;
992
993    switch (type) {
994    case TCG_TYPE_I32:
995        tcg_out_ldst(s, OPC_LW, arg, arg1, arg2);
996        break;
997    case TCG_TYPE_I64:
998        tcg_out_ldst(s, OPC_LD, arg, arg1, arg2);
999        break;
1000    case TCG_TYPE_V64:
1001    case TCG_TYPE_V128:
1002    case TCG_TYPE_V256:
1003        if (type >= riscv_lg2_vlenb) {
1004            static const RISCVInsn whole_reg_ld[] = {
1005                OPC_VL1RE64_V, OPC_VL2RE64_V, OPC_VL4RE64_V, OPC_VL8RE64_V
1006            };
1007            unsigned idx = type - riscv_lg2_vlenb;
1008
1009            tcg_debug_assert(idx < ARRAY_SIZE(whole_reg_ld));
1010            insn = whole_reg_ld[idx];
1011        } else {
1012            static const RISCVInsn unit_stride_ld[] = {
1013                OPC_VLE8_V, OPC_VLE16_V, OPC_VLE32_V, OPC_VLE64_V
1014            };
1015            MemOp prev_vsew = set_vtype_len(s, type);
1016
1017            tcg_debug_assert(prev_vsew < ARRAY_SIZE(unit_stride_ld));
1018            insn = unit_stride_ld[prev_vsew];
1019        }
1020        tcg_out_vec_ldst(s, insn, arg, arg1, arg2);
1021        break;
1022    default:
1023        g_assert_not_reached();
1024    }
1025}
1026
1027static void tcg_out_st(TCGContext *s, TCGType type, TCGReg arg,
1028                       TCGReg arg1, intptr_t arg2)
1029{
1030    RISCVInsn insn;
1031
1032    switch (type) {
1033    case TCG_TYPE_I32:
1034        tcg_out_ldst(s, OPC_SW, arg, arg1, arg2);
1035        break;
1036    case TCG_TYPE_I64:
1037        tcg_out_ldst(s, OPC_SD, arg, arg1, arg2);
1038        break;
1039    case TCG_TYPE_V64:
1040    case TCG_TYPE_V128:
1041    case TCG_TYPE_V256:
1042        if (type >= riscv_lg2_vlenb) {
1043            static const RISCVInsn whole_reg_st[] = {
1044                OPC_VS1R_V, OPC_VS2R_V, OPC_VS4R_V, OPC_VS8R_V
1045            };
1046            unsigned idx = type - riscv_lg2_vlenb;
1047
1048            tcg_debug_assert(idx < ARRAY_SIZE(whole_reg_st));
1049            insn = whole_reg_st[idx];
1050        } else {
1051            static const RISCVInsn unit_stride_st[] = {
1052                OPC_VSE8_V, OPC_VSE16_V, OPC_VSE32_V, OPC_VSE64_V
1053            };
1054            MemOp prev_vsew = set_vtype_len(s, type);
1055
1056            tcg_debug_assert(prev_vsew < ARRAY_SIZE(unit_stride_st));
1057            insn = unit_stride_st[prev_vsew];
1058        }
1059        tcg_out_vec_ldst(s, insn, arg, arg1, arg2);
1060        break;
1061    default:
1062        g_assert_not_reached();
1063    }
1064}
1065
1066static bool tcg_out_sti(TCGContext *s, TCGType type, TCGArg val,
1067                        TCGReg base, intptr_t ofs)
1068{
1069    if (val == 0) {
1070        tcg_out_st(s, type, TCG_REG_ZERO, base, ofs);
1071        return true;
1072    }
1073    return false;
1074}
1075
1076static bool tcg_out_dup_vec(TCGContext *s, TCGType type, unsigned vece,
1077                                   TCGReg dst, TCGReg src)
1078{
1079    set_vtype_len_sew(s, type, vece);
1080    tcg_out_opc_vx(s, OPC_VMV_V_X, dst, 0, src);
1081    return true;
1082}
1083
1084static bool tcg_out_dupm_vec(TCGContext *s, TCGType type, unsigned vece,
1085                                    TCGReg dst, TCGReg base, intptr_t offset)
1086{
1087    tcg_out_ld(s, TCG_TYPE_REG, TCG_REG_TMP0, base, offset);
1088    return tcg_out_dup_vec(s, type, vece, dst, TCG_REG_TMP0);
1089}
1090
1091static void tcg_out_dupi_vec(TCGContext *s, TCGType type, unsigned vece,
1092                                    TCGReg dst, int64_t arg)
1093{
1094    /* Arg is replicated by VECE; extract the highest element. */
1095    arg >>= (-8 << vece) & 63;
1096
1097    if (arg >= -16 && arg < 16) {
1098        if (arg == 0 || arg == -1) {
1099            set_vtype_len(s, type);
1100        } else {
1101            set_vtype_len_sew(s, type, vece);
1102        }
1103        tcg_out_opc_vi(s, OPC_VMV_V_I, dst, 0, arg);
1104        return;
1105    }
1106    tcg_out_movi(s, TCG_TYPE_I64, TCG_REG_TMP0, arg);
1107    tcg_out_dup_vec(s, type, vece, dst, TCG_REG_TMP0);
1108}
1109
1110static void tcg_out_br(TCGContext *s, TCGLabel *l)
1111{
1112    tcg_out_reloc(s, s->code_ptr, R_RISCV_JAL, l, 0);
1113    tcg_out_opc_jump(s, OPC_JAL, TCG_REG_ZERO, 0);
1114}
1115
1116static const struct {
1117    RISCVInsn op;
1118    bool swap;
1119} tcg_brcond_to_riscv[] = {
1120    [TCG_COND_EQ] =  { OPC_BEQ,  false },
1121    [TCG_COND_NE] =  { OPC_BNE,  false },
1122    [TCG_COND_LT] =  { OPC_BLT,  false },
1123    [TCG_COND_GE] =  { OPC_BGE,  false },
1124    [TCG_COND_LE] =  { OPC_BGE,  true  },
1125    [TCG_COND_GT] =  { OPC_BLT,  true  },
1126    [TCG_COND_LTU] = { OPC_BLTU, false },
1127    [TCG_COND_GEU] = { OPC_BGEU, false },
1128    [TCG_COND_LEU] = { OPC_BGEU, true  },
1129    [TCG_COND_GTU] = { OPC_BLTU, true  }
1130};
1131
1132static void tgen_brcond(TCGContext *s, TCGType type, TCGCond cond,
1133                        TCGReg arg1, TCGReg arg2, TCGLabel *l)
1134{
1135    RISCVInsn op = tcg_brcond_to_riscv[cond].op;
1136
1137    tcg_debug_assert(op != 0);
1138
1139    if (tcg_brcond_to_riscv[cond].swap) {
1140        TCGReg t = arg1;
1141        arg1 = arg2;
1142        arg2 = t;
1143    }
1144
1145    tcg_out_reloc(s, s->code_ptr, R_RISCV_BRANCH, l, 0);
1146    tcg_out_opc_branch(s, op, arg1, arg2, 0);
1147}
1148
1149static const TCGOutOpBrcond outop_brcond = {
1150    .base.static_constraint = C_O0_I2(r, rz),
1151    .out_rr = tgen_brcond,
1152};
1153
1154#define SETCOND_INV    TCG_TARGET_NB_REGS
1155#define SETCOND_NEZ    (SETCOND_INV << 1)
1156#define SETCOND_FLAGS  (SETCOND_INV | SETCOND_NEZ)
1157
1158static int tcg_out_setcond_int(TCGContext *s, TCGCond cond, TCGReg ret,
1159                               TCGReg arg1, tcg_target_long arg2, bool c2)
1160{
1161    int flags = 0;
1162
1163    switch (cond) {
1164    case TCG_COND_EQ:    /* -> NE  */
1165    case TCG_COND_GE:    /* -> LT  */
1166    case TCG_COND_GEU:   /* -> LTU */
1167    case TCG_COND_GT:    /* -> LE  */
1168    case TCG_COND_GTU:   /* -> LEU */
1169        cond = tcg_invert_cond(cond);
1170        flags ^= SETCOND_INV;
1171        break;
1172    default:
1173        break;
1174    }
1175
1176    switch (cond) {
1177    case TCG_COND_LE:
1178    case TCG_COND_LEU:
1179        /*
1180         * If we have a constant input, the most efficient way to implement
1181         * LE is by adding 1 and using LT.  Watch out for wrap around for LEU.
1182         * We don't need to care for this for LE because the constant input
1183         * is constrained to signed 12-bit, and 0x800 is representable in the
1184         * temporary register.
1185         */
1186        if (c2) {
1187            if (cond == TCG_COND_LEU) {
1188                /* unsigned <= -1 is true */
1189                if (arg2 == -1) {
1190                    tcg_out_movi(s, TCG_TYPE_REG, ret, !(flags & SETCOND_INV));
1191                    return ret;
1192                }
1193                cond = TCG_COND_LTU;
1194            } else {
1195                cond = TCG_COND_LT;
1196            }
1197            tcg_debug_assert(arg2 <= 0x7ff);
1198            if (++arg2 == 0x800) {
1199                tcg_out_movi(s, TCG_TYPE_REG, TCG_REG_TMP0, arg2);
1200                arg2 = TCG_REG_TMP0;
1201                c2 = false;
1202            }
1203        } else {
1204            TCGReg tmp = arg2;
1205            arg2 = arg1;
1206            arg1 = tmp;
1207            cond = tcg_swap_cond(cond);    /* LE -> GE */
1208            cond = tcg_invert_cond(cond);  /* GE -> LT */
1209            flags ^= SETCOND_INV;
1210        }
1211        break;
1212    default:
1213        break;
1214    }
1215
1216    switch (cond) {
1217    case TCG_COND_NE:
1218        flags |= SETCOND_NEZ;
1219        if (!c2) {
1220            tcg_out_opc_reg(s, OPC_XOR, ret, arg1, arg2);
1221        } else if (arg2 == 0) {
1222            ret = arg1;
1223        } else {
1224            tcg_out_opc_imm(s, OPC_XORI, ret, arg1, arg2);
1225        }
1226        break;
1227
1228    case TCG_COND_LT:
1229        if (c2) {
1230            tcg_out_opc_imm(s, OPC_SLTI, ret, arg1, arg2);
1231        } else {
1232            tcg_out_opc_reg(s, OPC_SLT, ret, arg1, arg2);
1233        }
1234        break;
1235
1236    case TCG_COND_LTU:
1237        if (c2) {
1238            tcg_out_opc_imm(s, OPC_SLTIU, ret, arg1, arg2);
1239        } else {
1240            tcg_out_opc_reg(s, OPC_SLTU, ret, arg1, arg2);
1241        }
1242        break;
1243
1244    default:
1245        g_assert_not_reached();
1246    }
1247
1248    return ret | flags;
1249}
1250
1251static void tcg_out_setcond(TCGContext *s, TCGCond cond, TCGReg ret,
1252                            TCGReg arg1, tcg_target_long arg2, bool c2)
1253{
1254    int tmpflags = tcg_out_setcond_int(s, cond, ret, arg1, arg2, c2);
1255
1256    if (tmpflags != ret) {
1257        TCGReg tmp = tmpflags & ~SETCOND_FLAGS;
1258
1259        switch (tmpflags & SETCOND_FLAGS) {
1260        case SETCOND_INV:
1261            /* Intermediate result is boolean: simply invert. */
1262            tcg_out_opc_imm(s, OPC_XORI, ret, tmp, 1);
1263            break;
1264        case SETCOND_NEZ:
1265            /* Intermediate result is zero/non-zero: test != 0. */
1266            tcg_out_opc_reg(s, OPC_SLTU, ret, TCG_REG_ZERO, tmp);
1267            break;
1268        case SETCOND_NEZ | SETCOND_INV:
1269            /* Intermediate result is zero/non-zero: test == 0. */
1270            tcg_out_opc_imm(s, OPC_SLTIU, ret, tmp, 1);
1271            break;
1272        default:
1273            g_assert_not_reached();
1274        }
1275    }
1276}
1277
1278static void tgen_setcond(TCGContext *s, TCGType type, TCGCond cond,
1279                         TCGReg dest, TCGReg arg1, TCGReg arg2)
1280{
1281    tcg_out_setcond(s, cond, dest, arg1, arg2, false);
1282}
1283
1284static void tgen_setcondi(TCGContext *s, TCGType type, TCGCond cond,
1285                          TCGReg dest, TCGReg arg1, tcg_target_long arg2)
1286{
1287    tcg_out_setcond(s, cond, dest, arg1, arg2, true);
1288}
1289
1290static const TCGOutOpSetcond outop_setcond = {
1291    .base.static_constraint = C_O1_I2(r, r, rI),
1292    .out_rrr = tgen_setcond,
1293    .out_rri = tgen_setcondi,
1294};
1295
1296static void tcg_out_negsetcond(TCGContext *s, TCGCond cond, TCGReg ret,
1297                               TCGReg arg1, tcg_target_long arg2, bool c2)
1298{
1299    int tmpflags;
1300    TCGReg tmp;
1301
1302    /* For LT/GE comparison against 0, replicate the sign bit. */
1303    if (c2 && arg2 == 0) {
1304        switch (cond) {
1305        case TCG_COND_GE:
1306            tcg_out_opc_imm(s, OPC_XORI, ret, arg1, -1);
1307            arg1 = ret;
1308            /* fall through */
1309        case TCG_COND_LT:
1310            tcg_out_opc_imm(s, OPC_SRAI, ret, arg1, TCG_TARGET_REG_BITS - 1);
1311            return;
1312        default:
1313            break;
1314        }
1315    }
1316
1317    tmpflags = tcg_out_setcond_int(s, cond, ret, arg1, arg2, c2);
1318    tmp = tmpflags & ~SETCOND_FLAGS;
1319
1320    /* If intermediate result is zero/non-zero: test != 0. */
1321    if (tmpflags & SETCOND_NEZ) {
1322        tcg_out_opc_reg(s, OPC_SLTU, ret, TCG_REG_ZERO, tmp);
1323        tmp = ret;
1324    }
1325
1326    /* Produce the 0/-1 result. */
1327    if (tmpflags & SETCOND_INV) {
1328        tcg_out_opc_imm(s, OPC_ADDI, ret, tmp, -1);
1329    } else {
1330        tcg_out_opc_reg(s, OPC_SUB, ret, TCG_REG_ZERO, tmp);
1331    }
1332}
1333
1334static void tgen_negsetcond(TCGContext *s, TCGType type, TCGCond cond,
1335                            TCGReg dest, TCGReg arg1, TCGReg arg2)
1336{
1337    tcg_out_negsetcond(s, cond, dest, arg1, arg2, false);
1338}
1339
1340static void tgen_negsetcondi(TCGContext *s, TCGType type, TCGCond cond,
1341                             TCGReg dest, TCGReg arg1, tcg_target_long arg2)
1342{
1343    tcg_out_negsetcond(s, cond, dest, arg1, arg2, true);
1344}
1345
1346static const TCGOutOpSetcond outop_negsetcond = {
1347    .base.static_constraint = C_O1_I2(r, r, rI),
1348    .out_rrr = tgen_negsetcond,
1349    .out_rri = tgen_negsetcondi,
1350};
1351
1352static void tcg_out_movcond_zicond(TCGContext *s, TCGReg ret, TCGReg test_ne,
1353                                   int val1, bool c_val1,
1354                                   int val2, bool c_val2)
1355{
1356    if (val1 == 0) {
1357        if (c_val2) {
1358            tcg_out_movi(s, TCG_TYPE_REG, TCG_REG_TMP1, val2);
1359            val2 = TCG_REG_TMP1;
1360        }
1361        tcg_out_opc_reg(s, OPC_CZERO_NEZ, ret, val2, test_ne);
1362        return;
1363    }
1364
1365    if (val2 == 0) {
1366        if (c_val1) {
1367            tcg_out_movi(s, TCG_TYPE_REG, TCG_REG_TMP1, val1);
1368            val1 = TCG_REG_TMP1;
1369        }
1370        tcg_out_opc_reg(s, OPC_CZERO_EQZ, ret, val1, test_ne);
1371        return;
1372    }
1373
1374    if (c_val2) {
1375        if (c_val1) {
1376            tcg_out_movi(s, TCG_TYPE_REG, TCG_REG_TMP1, val1 - val2);
1377        } else {
1378            tcg_out_opc_imm(s, OPC_ADDI, TCG_REG_TMP1, val1, -val2);
1379        }
1380        tcg_out_opc_reg(s, OPC_CZERO_EQZ, ret, TCG_REG_TMP1, test_ne);
1381        tcg_out_opc_imm(s, OPC_ADDI, ret, ret, val2);
1382        return;
1383    }
1384
1385    if (c_val1) {
1386        tcg_out_opc_imm(s, OPC_ADDI, TCG_REG_TMP1, val2, -val1);
1387        tcg_out_opc_reg(s, OPC_CZERO_NEZ, ret, TCG_REG_TMP1, test_ne);
1388        tcg_out_opc_imm(s, OPC_ADDI, ret, ret, val1);
1389        return;
1390    }
1391
1392    tcg_out_opc_reg(s, OPC_CZERO_NEZ, TCG_REG_TMP1, val2, test_ne);
1393    tcg_out_opc_reg(s, OPC_CZERO_EQZ, TCG_REG_TMP0, val1, test_ne);
1394    tcg_out_opc_reg(s, OPC_OR, ret, TCG_REG_TMP0, TCG_REG_TMP1);
1395}
1396
1397static void tcg_out_movcond_br1(TCGContext *s, TCGCond cond, TCGReg ret,
1398                                TCGReg cmp1, TCGReg cmp2,
1399                                int val, bool c_val)
1400{
1401    RISCVInsn op;
1402    int disp = 8;
1403
1404    tcg_debug_assert((unsigned)cond < ARRAY_SIZE(tcg_brcond_to_riscv));
1405    op = tcg_brcond_to_riscv[cond].op;
1406    tcg_debug_assert(op != 0);
1407
1408    if (tcg_brcond_to_riscv[cond].swap) {
1409        tcg_out_opc_branch(s, op, cmp2, cmp1, disp);
1410    } else {
1411        tcg_out_opc_branch(s, op, cmp1, cmp2, disp);
1412    }
1413    if (c_val) {
1414        tcg_out_opc_imm(s, OPC_ADDI, ret, TCG_REG_ZERO, val);
1415    } else {
1416        tcg_out_opc_imm(s, OPC_ADDI, ret, val, 0);
1417    }
1418}
1419
1420static void tcg_out_movcond_br2(TCGContext *s, TCGCond cond, TCGReg ret,
1421                                TCGReg cmp1, TCGReg cmp2,
1422                                int val1, bool c_val1,
1423                                int val2, bool c_val2)
1424{
1425    TCGReg tmp;
1426
1427    /* TCG optimizer reorders to prefer ret matching val2. */
1428    if (!c_val2 && ret == val2) {
1429        cond = tcg_invert_cond(cond);
1430        tcg_out_movcond_br1(s, cond, ret, cmp1, cmp2, val1, c_val1);
1431        return;
1432    }
1433
1434    if (!c_val1 && ret == val1) {
1435        tcg_out_movcond_br1(s, cond, ret, cmp1, cmp2, val2, c_val2);
1436        return;
1437    }
1438
1439    tmp = (ret == cmp1 || ret == cmp2 ? TCG_REG_TMP1 : ret);
1440    if (c_val1) {
1441        tcg_out_movi(s, TCG_TYPE_REG, tmp, val1);
1442    } else {
1443        tcg_out_mov(s, TCG_TYPE_REG, tmp, val1);
1444    }
1445    tcg_out_movcond_br1(s, cond, tmp, cmp1, cmp2, val2, c_val2);
1446    tcg_out_mov(s, TCG_TYPE_REG, ret, tmp);
1447}
1448
1449static void tcg_out_movcond(TCGContext *s, TCGType type, TCGCond cond,
1450                            TCGReg ret, TCGReg cmp1, TCGArg cmp2, bool c_cmp2,
1451                            TCGArg val1, bool c_val1,
1452                            TCGArg val2, bool c_val2)
1453{
1454    int tmpflags;
1455    TCGReg t;
1456
1457    if (!(cpuinfo & CPUINFO_ZICOND) && (!c_cmp2 || cmp2 == 0)) {
1458        tcg_out_movcond_br2(s, cond, ret, cmp1, cmp2,
1459                            val1, c_val1, val2, c_val2);
1460        return;
1461    }
1462
1463    tmpflags = tcg_out_setcond_int(s, cond, TCG_REG_TMP0, cmp1, cmp2, c_cmp2);
1464    t = tmpflags & ~SETCOND_FLAGS;
1465
1466    if (cpuinfo & CPUINFO_ZICOND) {
1467        if (tmpflags & SETCOND_INV) {
1468            tcg_out_movcond_zicond(s, ret, t, val2, c_val2, val1, c_val1);
1469        } else {
1470            tcg_out_movcond_zicond(s, ret, t, val1, c_val1, val2, c_val2);
1471        }
1472    } else {
1473        cond = tmpflags & SETCOND_INV ? TCG_COND_EQ : TCG_COND_NE;
1474        tcg_out_movcond_br2(s, cond, ret, t, TCG_REG_ZERO,
1475                            val1, c_val1, val2, c_val2);
1476    }
1477}
1478
1479static const TCGOutOpMovcond outop_movcond = {
1480    .base.static_constraint = C_O1_I4(r, r, rI, rM, rM),
1481    .out = tcg_out_movcond,
1482};
1483
1484static void tcg_out_cltz(TCGContext *s, TCGType type, RISCVInsn insn,
1485                         TCGReg ret, TCGReg src1, int src2, bool c_src2)
1486{
1487    tcg_out_opc_imm(s, insn, ret, src1, 0);
1488
1489    if (!c_src2 || src2 != (type == TCG_TYPE_I32 ? 32 : 64)) {
1490        /*
1491         * The requested zero result does not match the insn, so adjust.
1492         * Note that constraints put 'ret' in a new register, so the
1493         * computation above did not clobber either 'src1' or 'src2'.
1494         */
1495        tcg_out_movcond(s, type, TCG_COND_EQ, ret, src1, 0, true,
1496                        src2, c_src2, ret, false);
1497    }
1498}
1499
1500static void tcg_out_cmpsel(TCGContext *s, TCGType type, unsigned vece,
1501                           TCGCond cond, TCGReg ret,
1502                           TCGReg cmp1, TCGReg cmp2, bool c_cmp2,
1503                           TCGReg val1, bool c_val1,
1504                           TCGReg val2, bool c_val2)
1505{
1506    set_vtype_len_sew(s, type, vece);
1507
1508    /* Use only vmerge_vim if possible, by inverting the test. */
1509    if (c_val2 && !c_val1) {
1510        TCGArg temp = val1;
1511        cond = tcg_invert_cond(cond);
1512        val1 = val2;
1513        val2 = temp;
1514        c_val1 = true;
1515        c_val2 = false;
1516    }
1517
1518    /* Perform the comparison into V0 mask. */
1519    if (c_cmp2) {
1520        tcg_out_opc_vi(s, tcg_cmpcond_to_rvv_vi[cond].op, TCG_REG_V0, cmp1,
1521                       cmp2 - tcg_cmpcond_to_rvv_vi[cond].adjust);
1522    } else if (tcg_cmpcond_to_rvv_vv[cond].swap) {
1523        tcg_out_opc_vv(s, tcg_cmpcond_to_rvv_vv[cond].op,
1524                       TCG_REG_V0, cmp2, cmp1);
1525    } else {
1526        tcg_out_opc_vv(s, tcg_cmpcond_to_rvv_vv[cond].op,
1527                       TCG_REG_V0, cmp1, cmp2);
1528    }
1529    if (c_val1) {
1530        if (c_val2) {
1531            tcg_out_opc_vi(s, OPC_VMV_V_I, ret, 0, val2);
1532            val2 = ret;
1533        }
1534        /* vd[i] == v0.mask[i] ? imm : vs2[i] */
1535        tcg_out_opc_vim_mask(s, OPC_VMERGE_VIM, ret, val2, val1);
1536    } else {
1537        /* vd[i] == v0.mask[i] ? vs1[i] : vs2[i] */
1538        tcg_out_opc_vvm_mask(s, OPC_VMERGE_VVM, ret, val2, val1);
1539    }
1540}
1541
1542static void tcg_out_vshifti(TCGContext *s, RISCVInsn opc_vi, RISCVInsn opc_vx,
1543                             TCGReg dst, TCGReg src, unsigned imm)
1544{
1545    if (imm < 32) {
1546        tcg_out_opc_vi(s, opc_vi, dst, src, imm);
1547    } else {
1548        tcg_out_movi(s, TCG_TYPE_I32, TCG_REG_TMP0, imm);
1549        tcg_out_opc_vx(s, opc_vx, dst, src, TCG_REG_TMP0);
1550    }
1551}
1552
1553static void init_setting_vtype(TCGContext *s)
1554{
1555    s->riscv_cur_type = TCG_TYPE_COUNT;
1556}
1557
1558static void tcg_out_call_int(TCGContext *s, const tcg_insn_unit *arg, bool tail)
1559{
1560    TCGReg link = tail ? TCG_REG_ZERO : TCG_REG_RA;
1561    ptrdiff_t offset = tcg_pcrel_diff(s, arg);
1562    int ret;
1563
1564    init_setting_vtype(s);
1565
1566    tcg_debug_assert((offset & 1) == 0);
1567    if (offset == sextreg(offset, 0, 20)) {
1568        /* short jump: -2097150 to 2097152 */
1569        tcg_out_opc_jump(s, OPC_JAL, link, offset);
1570    } else if (offset == (int32_t)offset) {
1571        /* long jump: -2147483646 to 2147483648 */
1572        tcg_out_opc_upper(s, OPC_AUIPC, TCG_REG_TMP0, 0);
1573        tcg_out_opc_imm(s, OPC_JALR, link, TCG_REG_TMP0, 0);
1574        ret = reloc_call(s->code_ptr - 2, arg);
1575        tcg_debug_assert(ret == true);
1576    } else {
1577        /* far jump: 64-bit */
1578        tcg_target_long imm = sextreg((tcg_target_long)arg, 0, 12);
1579        tcg_target_long base = (tcg_target_long)arg - imm;
1580        tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_TMP0, base);
1581        tcg_out_opc_imm(s, OPC_JALR, link, TCG_REG_TMP0, imm);
1582    }
1583}
1584
1585static void tcg_out_call(TCGContext *s, const tcg_insn_unit *arg,
1586                         const TCGHelperInfo *info)
1587{
1588    tcg_out_call_int(s, arg, false);
1589}
1590
1591static void tcg_out_mb(TCGContext *s, unsigned a0)
1592{
1593    tcg_insn_unit insn = OPC_FENCE;
1594
1595    if (a0 & TCG_MO_LD_LD) {
1596        insn |= 0x02200000;
1597    }
1598    if (a0 & TCG_MO_ST_LD) {
1599        insn |= 0x01200000;
1600    }
1601    if (a0 & TCG_MO_LD_ST) {
1602        insn |= 0x02100000;
1603    }
1604    if (a0 & TCG_MO_ST_ST) {
1605        insn |= 0x01100000;
1606    }
1607    tcg_out32(s, insn);
1608}
1609
1610/*
1611 * Load/store and TLB
1612 */
1613
1614static void tcg_out_goto(TCGContext *s, const tcg_insn_unit *target)
1615{
1616    tcg_out_opc_jump(s, OPC_JAL, TCG_REG_ZERO, 0);
1617    bool ok = reloc_jimm20(s->code_ptr - 1, target);
1618    tcg_debug_assert(ok);
1619}
1620
1621bool tcg_target_has_memory_bswap(MemOp memop)
1622{
1623    return false;
1624}
1625
1626/* We have three temps, we might as well expose them. */
1627static const TCGLdstHelperParam ldst_helper_param = {
1628    .ntmp = 3, .tmp = { TCG_REG_TMP0, TCG_REG_TMP1, TCG_REG_TMP2 }
1629};
1630
1631static bool tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *l)
1632{
1633    MemOp opc = get_memop(l->oi);
1634
1635    /* resolve label address */
1636    if (!reloc_sbimm12(l->label_ptr[0], tcg_splitwx_to_rx(s->code_ptr))) {
1637        return false;
1638    }
1639
1640    /* call load helper */
1641    tcg_out_ld_helper_args(s, l, &ldst_helper_param);
1642    tcg_out_call_int(s, qemu_ld_helpers[opc & MO_SSIZE], false);
1643    tcg_out_ld_helper_ret(s, l, true, &ldst_helper_param);
1644
1645    tcg_out_goto(s, l->raddr);
1646    return true;
1647}
1648
1649static bool tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *l)
1650{
1651    MemOp opc = get_memop(l->oi);
1652
1653    /* resolve label address */
1654    if (!reloc_sbimm12(l->label_ptr[0], tcg_splitwx_to_rx(s->code_ptr))) {
1655        return false;
1656    }
1657
1658    /* call store helper */
1659    tcg_out_st_helper_args(s, l, &ldst_helper_param);
1660    tcg_out_call_int(s, qemu_st_helpers[opc & MO_SIZE], false);
1661
1662    tcg_out_goto(s, l->raddr);
1663    return true;
1664}
1665
1666/* We expect to use a 12-bit negative offset from ENV.  */
1667#define MIN_TLB_MASK_TABLE_OFS  -(1 << 11)
1668
1669/*
1670 * For system-mode, perform the TLB load and compare.
1671 * For user-mode, perform any required alignment tests.
1672 * In both cases, return a TCGLabelQemuLdst structure if the slow path
1673 * is required and fill in @h with the host address for the fast path.
1674 */
1675static TCGLabelQemuLdst *prepare_host_addr(TCGContext *s, TCGReg *pbase,
1676                                           TCGReg addr_reg, MemOpIdx oi,
1677                                           bool is_ld)
1678{
1679    TCGType addr_type = s->addr_type;
1680    TCGLabelQemuLdst *ldst = NULL;
1681    MemOp opc = get_memop(oi);
1682    TCGAtomAlign aa;
1683    unsigned a_mask;
1684
1685    aa = atom_and_align_for_opc(s, opc, MO_ATOM_IFALIGN, false);
1686    a_mask = (1u << aa.align) - 1;
1687
1688    if (tcg_use_softmmu) {
1689        unsigned s_bits = opc & MO_SIZE;
1690        unsigned s_mask = (1u << s_bits) - 1;
1691        int mem_index = get_mmuidx(oi);
1692        int fast_ofs = tlb_mask_table_ofs(s, mem_index);
1693        int mask_ofs = fast_ofs + offsetof(CPUTLBDescFast, mask);
1694        int table_ofs = fast_ofs + offsetof(CPUTLBDescFast, table);
1695        int compare_mask;
1696        TCGReg addr_adj;
1697
1698        ldst = new_ldst_label(s);
1699        ldst->is_ld = is_ld;
1700        ldst->oi = oi;
1701        ldst->addr_reg = addr_reg;
1702
1703        init_setting_vtype(s);
1704
1705        tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_TMP0, TCG_AREG0, mask_ofs);
1706        tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_TMP1, TCG_AREG0, table_ofs);
1707
1708        tcg_out_opc_imm(s, OPC_SRLI, TCG_REG_TMP2, addr_reg,
1709                        TARGET_PAGE_BITS - CPU_TLB_ENTRY_BITS);
1710        tcg_out_opc_reg(s, OPC_AND, TCG_REG_TMP2, TCG_REG_TMP2, TCG_REG_TMP0);
1711        tcg_out_opc_reg(s, OPC_ADD, TCG_REG_TMP2, TCG_REG_TMP2, TCG_REG_TMP1);
1712
1713        /*
1714         * For aligned accesses, we check the first byte and include the
1715         * alignment bits within the address.  For unaligned access, we
1716         * check that we don't cross pages using the address of the last
1717         * byte of the access.
1718         */
1719        addr_adj = addr_reg;
1720        if (a_mask < s_mask) {
1721            addr_adj = TCG_REG_TMP0;
1722            tcg_out_opc_imm(s, addr_type == TCG_TYPE_I32 ? OPC_ADDIW : OPC_ADDI,
1723                            addr_adj, addr_reg, s_mask - a_mask);
1724        }
1725        compare_mask = TARGET_PAGE_MASK | a_mask;
1726        if (compare_mask == sextreg(compare_mask, 0, 12)) {
1727            tcg_out_opc_imm(s, OPC_ANDI, TCG_REG_TMP1, addr_adj, compare_mask);
1728        } else {
1729            tcg_out_movi(s, addr_type, TCG_REG_TMP1, compare_mask);
1730            tcg_out_opc_reg(s, OPC_AND, TCG_REG_TMP1, TCG_REG_TMP1, addr_adj);
1731        }
1732
1733        /* Load the tlb comparator and the addend.  */
1734        QEMU_BUILD_BUG_ON(HOST_BIG_ENDIAN);
1735        tcg_out_ld(s, addr_type, TCG_REG_TMP0, TCG_REG_TMP2,
1736                   is_ld ? offsetof(CPUTLBEntry, addr_read)
1737                         : offsetof(CPUTLBEntry, addr_write));
1738        tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_TMP2, TCG_REG_TMP2,
1739                   offsetof(CPUTLBEntry, addend));
1740
1741        /* Compare masked address with the TLB entry. */
1742        ldst->label_ptr[0] = s->code_ptr;
1743        tcg_out_opc_branch(s, OPC_BNE, TCG_REG_TMP0, TCG_REG_TMP1, 0);
1744
1745        /* TLB Hit - translate address using addend.  */
1746        if (addr_type != TCG_TYPE_I32) {
1747            tcg_out_opc_reg(s, OPC_ADD, TCG_REG_TMP0, addr_reg, TCG_REG_TMP2);
1748        } else if (cpuinfo & CPUINFO_ZBA) {
1749            tcg_out_opc_reg(s, OPC_ADD_UW, TCG_REG_TMP0,
1750                            addr_reg, TCG_REG_TMP2);
1751        } else {
1752            tcg_out_ext32u(s, TCG_REG_TMP0, addr_reg);
1753            tcg_out_opc_reg(s, OPC_ADD, TCG_REG_TMP0,
1754                            TCG_REG_TMP0, TCG_REG_TMP2);
1755        }
1756        *pbase = TCG_REG_TMP0;
1757    } else {
1758        TCGReg base;
1759
1760        if (a_mask) {
1761            ldst = new_ldst_label(s);
1762            ldst->is_ld = is_ld;
1763            ldst->oi = oi;
1764            ldst->addr_reg = addr_reg;
1765
1766            init_setting_vtype(s);
1767
1768            /* We are expecting alignment max 7, so we can always use andi. */
1769            tcg_debug_assert(a_mask == sextreg(a_mask, 0, 12));
1770            tcg_out_opc_imm(s, OPC_ANDI, TCG_REG_TMP1, addr_reg, a_mask);
1771
1772            ldst->label_ptr[0] = s->code_ptr;
1773            tcg_out_opc_branch(s, OPC_BNE, TCG_REG_TMP1, TCG_REG_ZERO, 0);
1774        }
1775
1776        if (guest_base != 0) {
1777            base = TCG_REG_TMP0;
1778            if (addr_type != TCG_TYPE_I32) {
1779                tcg_out_opc_reg(s, OPC_ADD, base, addr_reg,
1780                                TCG_GUEST_BASE_REG);
1781            } else if (cpuinfo & CPUINFO_ZBA) {
1782                tcg_out_opc_reg(s, OPC_ADD_UW, base, addr_reg,
1783                                TCG_GUEST_BASE_REG);
1784            } else {
1785                tcg_out_ext32u(s, base, addr_reg);
1786                tcg_out_opc_reg(s, OPC_ADD, base, base, TCG_GUEST_BASE_REG);
1787            }
1788        } else if (addr_type != TCG_TYPE_I32) {
1789            base = addr_reg;
1790        } else {
1791            base = TCG_REG_TMP0;
1792            tcg_out_ext32u(s, base, addr_reg);
1793        }
1794        *pbase = base;
1795    }
1796
1797    return ldst;
1798}
1799
1800static void tcg_out_qemu_ld_direct(TCGContext *s, TCGReg val,
1801                                   TCGReg base, MemOp opc, TCGType type)
1802{
1803    /* Byte swapping is left to middle-end expansion. */
1804    tcg_debug_assert((opc & MO_BSWAP) == 0);
1805
1806    switch (opc & (MO_SSIZE)) {
1807    case MO_UB:
1808        tcg_out_opc_imm(s, OPC_LBU, val, base, 0);
1809        break;
1810    case MO_SB:
1811        tcg_out_opc_imm(s, OPC_LB, val, base, 0);
1812        break;
1813    case MO_UW:
1814        tcg_out_opc_imm(s, OPC_LHU, val, base, 0);
1815        break;
1816    case MO_SW:
1817        tcg_out_opc_imm(s, OPC_LH, val, base, 0);
1818        break;
1819    case MO_UL:
1820        if (type == TCG_TYPE_I64) {
1821            tcg_out_opc_imm(s, OPC_LWU, val, base, 0);
1822            break;
1823        }
1824        /* FALLTHRU */
1825    case MO_SL:
1826        tcg_out_opc_imm(s, OPC_LW, val, base, 0);
1827        break;
1828    case MO_UQ:
1829        tcg_out_opc_imm(s, OPC_LD, val, base, 0);
1830        break;
1831    default:
1832        g_assert_not_reached();
1833    }
1834}
1835
1836static void tgen_qemu_ld(TCGContext *s, TCGType type, TCGReg data_reg,
1837                         TCGReg addr_reg, MemOpIdx oi)
1838{
1839    TCGLabelQemuLdst *ldst;
1840    TCGReg base;
1841
1842    ldst = prepare_host_addr(s, &base, addr_reg, oi, true);
1843    tcg_out_qemu_ld_direct(s, data_reg, base, get_memop(oi), type);
1844
1845    if (ldst) {
1846        ldst->type = type;
1847        ldst->datalo_reg = data_reg;
1848        ldst->raddr = tcg_splitwx_to_rx(s->code_ptr);
1849    }
1850}
1851
1852static const TCGOutOpQemuLdSt outop_qemu_ld = {
1853    .base.static_constraint = C_O1_I1(r, r),
1854    .out = tgen_qemu_ld,
1855};
1856
1857static const TCGOutOpQemuLdSt2 outop_qemu_ld2 = {
1858    .base.static_constraint = C_NotImplemented,
1859};
1860
1861static void tcg_out_qemu_st_direct(TCGContext *s, TCGReg val,
1862                                   TCGReg base, MemOp opc)
1863{
1864    /* Byte swapping is left to middle-end expansion. */
1865    tcg_debug_assert((opc & MO_BSWAP) == 0);
1866
1867    switch (opc & (MO_SSIZE)) {
1868    case MO_8:
1869        tcg_out_opc_store(s, OPC_SB, base, val, 0);
1870        break;
1871    case MO_16:
1872        tcg_out_opc_store(s, OPC_SH, base, val, 0);
1873        break;
1874    case MO_32:
1875        tcg_out_opc_store(s, OPC_SW, base, val, 0);
1876        break;
1877    case MO_64:
1878        tcg_out_opc_store(s, OPC_SD, base, val, 0);
1879        break;
1880    default:
1881        g_assert_not_reached();
1882    }
1883}
1884
1885static void tgen_qemu_st(TCGContext *s, TCGType type, TCGReg data_reg,
1886                         TCGReg addr_reg, MemOpIdx oi)
1887{
1888    TCGLabelQemuLdst *ldst;
1889    TCGReg base;
1890
1891    ldst = prepare_host_addr(s, &base, addr_reg, oi, false);
1892    tcg_out_qemu_st_direct(s, data_reg, base, get_memop(oi));
1893
1894    if (ldst) {
1895        ldst->type = type;
1896        ldst->datalo_reg = data_reg;
1897        ldst->raddr = tcg_splitwx_to_rx(s->code_ptr);
1898    }
1899}
1900
1901static const TCGOutOpQemuLdSt outop_qemu_st = {
1902    .base.static_constraint = C_O0_I2(rz, r),
1903    .out = tgen_qemu_st,
1904};
1905
1906static const TCGOutOpQemuLdSt2 outop_qemu_st2 = {
1907    .base.static_constraint = C_NotImplemented,
1908};
1909
1910static const tcg_insn_unit *tb_ret_addr;
1911
1912static void tcg_out_exit_tb(TCGContext *s, uintptr_t a0)
1913{
1914    /* Reuse the zeroing that exists for goto_ptr.  */
1915    if (a0 == 0) {
1916        tcg_out_call_int(s, tcg_code_gen_epilogue, true);
1917    } else {
1918        tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_A0, a0);
1919        tcg_out_call_int(s, tb_ret_addr, true);
1920    }
1921}
1922
1923static void tcg_out_goto_tb(TCGContext *s, int which)
1924{
1925    /* Direct branch will be patched by tb_target_set_jmp_target. */
1926    set_jmp_insn_offset(s, which);
1927    tcg_out32(s, OPC_JAL);
1928
1929    /* When branch is out of range, fall through to indirect. */
1930    tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_TMP0, TCG_REG_ZERO,
1931               get_jmp_target_addr(s, which));
1932    tcg_out_opc_imm(s, OPC_JALR, TCG_REG_ZERO, TCG_REG_TMP0, 0);
1933    set_jmp_reset_offset(s, which);
1934}
1935
1936static void tcg_out_goto_ptr(TCGContext *s, TCGReg a0)
1937{
1938    tcg_out_opc_imm(s, OPC_JALR, TCG_REG_ZERO, a0, 0);
1939}
1940
1941void tb_target_set_jmp_target(const TranslationBlock *tb, int n,
1942                              uintptr_t jmp_rx, uintptr_t jmp_rw)
1943{
1944    uintptr_t addr = tb->jmp_target_addr[n];
1945    ptrdiff_t offset = addr - jmp_rx;
1946    tcg_insn_unit insn;
1947
1948    /* Either directly branch, or fall through to indirect branch. */
1949    if (offset == sextreg(offset, 0, 20)) {
1950        insn = encode_uj(OPC_JAL, TCG_REG_ZERO, offset);
1951    } else {
1952        insn = OPC_NOP;
1953    }
1954    qatomic_set((uint32_t *)jmp_rw, insn);
1955    flush_idcache_range(jmp_rx, jmp_rw, 4);
1956}
1957
1958
1959static void tgen_add(TCGContext *s, TCGType type,
1960                     TCGReg a0, TCGReg a1, TCGReg a2)
1961{
1962    RISCVInsn insn = type == TCG_TYPE_I32 ? OPC_ADDW : OPC_ADD;
1963    tcg_out_opc_reg(s, insn, a0, a1, a2);
1964}
1965
1966static void tgen_addi(TCGContext *s, TCGType type,
1967                      TCGReg a0, TCGReg a1, tcg_target_long a2)
1968{
1969    RISCVInsn insn = type == TCG_TYPE_I32 ? OPC_ADDIW : OPC_ADDI;
1970    tcg_out_opc_imm(s, insn, a0, a1, a2);
1971}
1972
1973static const TCGOutOpBinary outop_add = {
1974    .base.static_constraint = C_O1_I2(r, r, rI),
1975    .out_rrr = tgen_add,
1976    .out_rri = tgen_addi,
1977};
1978
1979static const TCGOutOpBinary outop_addco = {
1980    .base.static_constraint = C_NotImplemented,
1981};
1982
1983static const TCGOutOpAddSubCarry outop_addci = {
1984    .base.static_constraint = C_NotImplemented,
1985};
1986
1987static const TCGOutOpBinary outop_addcio = {
1988    .base.static_constraint = C_NotImplemented,
1989};
1990
1991static void tcg_out_set_carry(TCGContext *s)
1992{
1993    g_assert_not_reached();
1994}
1995
1996static void tgen_and(TCGContext *s, TCGType type,
1997                     TCGReg a0, TCGReg a1, TCGReg a2)
1998{
1999    tcg_out_opc_reg(s, OPC_AND, a0, a1, a2);
2000}
2001
2002static void tgen_andi(TCGContext *s, TCGType type,
2003                      TCGReg a0, TCGReg a1, tcg_target_long a2)
2004{
2005    tcg_out_opc_imm(s, OPC_ANDI, a0, a1, a2);
2006}
2007
2008static const TCGOutOpBinary outop_and = {
2009    .base.static_constraint = C_O1_I2(r, r, rI),
2010    .out_rrr = tgen_and,
2011    .out_rri = tgen_andi,
2012};
2013
2014static void tgen_andc(TCGContext *s, TCGType type,
2015                      TCGReg a0, TCGReg a1, TCGReg a2)
2016{
2017    tcg_out_opc_reg(s, OPC_ANDN, a0, a1, a2);
2018}
2019
2020static TCGConstraintSetIndex cset_zbb_rrr(TCGType type, unsigned flags)
2021{
2022    return cpuinfo & CPUINFO_ZBB ? C_O1_I2(r, r, r) : C_NotImplemented;
2023}
2024
2025static const TCGOutOpBinary outop_andc = {
2026    .base.static_constraint = C_Dynamic,
2027    .base.dynamic_constraint = cset_zbb_rrr,
2028    .out_rrr = tgen_andc,
2029};
2030
2031static void tgen_clz(TCGContext *s, TCGType type,
2032                     TCGReg a0, TCGReg a1, TCGReg a2)
2033{
2034    RISCVInsn insn = type == TCG_TYPE_I32 ? OPC_CLZW : OPC_CLZ;
2035    tcg_out_cltz(s, type, insn, a0, a1, a2, false);
2036}
2037
2038static void tgen_clzi(TCGContext *s, TCGType type,
2039                      TCGReg a0, TCGReg a1, tcg_target_long a2)
2040{
2041    RISCVInsn insn = type == TCG_TYPE_I32 ? OPC_CLZW : OPC_CLZ;
2042    tcg_out_cltz(s, type, insn, a0, a1, a2, true);
2043}
2044
2045static TCGConstraintSetIndex cset_clzctz(TCGType type, unsigned flags)
2046{
2047    return cpuinfo & CPUINFO_ZBB ? C_N1_I2(r, r, rM) : C_NotImplemented;
2048}
2049
2050static const TCGOutOpBinary outop_clz = {
2051    .base.static_constraint = C_Dynamic,
2052    .base.dynamic_constraint = cset_clzctz,
2053    .out_rrr = tgen_clz,
2054    .out_rri = tgen_clzi,
2055};
2056
2057static void tgen_ctpop(TCGContext *s, TCGType type, TCGReg a0, TCGReg a1)
2058{
2059    RISCVInsn insn = type == TCG_TYPE_I32 ? OPC_CPOPW : OPC_CPOP;
2060    tcg_out_opc_imm(s, insn, a0, a1, 0);
2061}
2062
2063static TCGConstraintSetIndex cset_ctpop(TCGType type, unsigned flags)
2064{
2065    return cpuinfo & CPUINFO_ZBB ? C_O1_I1(r, r) : C_NotImplemented;
2066}
2067
2068static const TCGOutOpUnary outop_ctpop = {
2069    .base.static_constraint = C_Dynamic,
2070    .base.dynamic_constraint = cset_ctpop,
2071    .out_rr = tgen_ctpop,
2072};
2073
2074static void tgen_ctz(TCGContext *s, TCGType type,
2075                     TCGReg a0, TCGReg a1, TCGReg a2)
2076{
2077    RISCVInsn insn = type == TCG_TYPE_I32 ? OPC_CTZW : OPC_CTZ;
2078    tcg_out_cltz(s, type, insn, a0, a1, a2, false);
2079}
2080
2081static void tgen_ctzi(TCGContext *s, TCGType type,
2082                      TCGReg a0, TCGReg a1, tcg_target_long a2)
2083{
2084    RISCVInsn insn = type == TCG_TYPE_I32 ? OPC_CTZW : OPC_CTZ;
2085    tcg_out_cltz(s, type, insn, a0, a1, a2, true);
2086}
2087
2088static const TCGOutOpBinary outop_ctz = {
2089    .base.static_constraint = C_Dynamic,
2090    .base.dynamic_constraint = cset_clzctz,
2091    .out_rrr = tgen_ctz,
2092    .out_rri = tgen_ctzi,
2093};
2094
2095static void tgen_divs(TCGContext *s, TCGType type,
2096                      TCGReg a0, TCGReg a1, TCGReg a2)
2097{
2098    RISCVInsn insn = type == TCG_TYPE_I32 ? OPC_DIVW : OPC_DIV;
2099    tcg_out_opc_reg(s, insn, a0, a1, a2);
2100}
2101
2102static const TCGOutOpBinary outop_divs = {
2103    .base.static_constraint = C_O1_I2(r, r, r),
2104    .out_rrr = tgen_divs,
2105};
2106
2107static const TCGOutOpDivRem outop_divs2 = {
2108    .base.static_constraint = C_NotImplemented,
2109};
2110
2111static void tgen_divu(TCGContext *s, TCGType type,
2112                      TCGReg a0, TCGReg a1, TCGReg a2)
2113{
2114    RISCVInsn insn = type == TCG_TYPE_I32 ? OPC_DIVUW : OPC_DIVU;
2115    tcg_out_opc_reg(s, insn, a0, a1, a2);
2116}
2117
2118static const TCGOutOpBinary outop_divu = {
2119    .base.static_constraint = C_O1_I2(r, r, r),
2120    .out_rrr = tgen_divu,
2121};
2122
2123static const TCGOutOpDivRem outop_divu2 = {
2124    .base.static_constraint = C_NotImplemented,
2125};
2126
2127static void tgen_eqv(TCGContext *s, TCGType type,
2128                     TCGReg a0, TCGReg a1, TCGReg a2)
2129{
2130    tcg_out_opc_reg(s, OPC_XNOR, a0, a1, a2);
2131}
2132
2133static const TCGOutOpBinary outop_eqv = {
2134    .base.static_constraint = C_Dynamic,
2135    .base.dynamic_constraint = cset_zbb_rrr,
2136    .out_rrr = tgen_eqv,
2137};
2138
2139static void tgen_extrh_i64_i32(TCGContext *s, TCGType t, TCGReg a0, TCGReg a1)
2140{
2141    tcg_out_opc_imm(s, OPC_SRAI, a0, a1, 32);
2142}
2143
2144static const TCGOutOpUnary outop_extrh_i64_i32 = {
2145    .base.static_constraint = C_O1_I1(r, r),
2146    .out_rr = tgen_extrh_i64_i32,
2147};
2148
2149static void tgen_mul(TCGContext *s, TCGType type,
2150                     TCGReg a0, TCGReg a1, TCGReg a2)
2151{
2152    RISCVInsn insn = type == TCG_TYPE_I32 ? OPC_MULW : OPC_MUL;
2153    tcg_out_opc_reg(s, insn, a0, a1, a2);
2154}
2155
2156static const TCGOutOpBinary outop_mul = {
2157    .base.static_constraint = C_O1_I2(r, r, r),
2158    .out_rrr = tgen_mul,
2159};
2160
2161static const TCGOutOpMul2 outop_muls2 = {
2162    .base.static_constraint = C_NotImplemented,
2163};
2164
2165static TCGConstraintSetIndex cset_mulh(TCGType type, unsigned flags)
2166{
2167    return type == TCG_TYPE_I32 ? C_NotImplemented : C_O1_I2(r, r, r);
2168}
2169
2170static void tgen_mulsh(TCGContext *s, TCGType type,
2171                       TCGReg a0, TCGReg a1, TCGReg a2)
2172{
2173    tcg_out_opc_reg(s, OPC_MULH, a0, a1, a2);
2174}
2175
2176static const TCGOutOpBinary outop_mulsh = {
2177    .base.static_constraint = C_Dynamic,
2178    .base.dynamic_constraint = cset_mulh,
2179    .out_rrr = tgen_mulsh,
2180};
2181
2182static const TCGOutOpMul2 outop_mulu2 = {
2183    .base.static_constraint = C_NotImplemented,
2184};
2185
2186static void tgen_muluh(TCGContext *s, TCGType type,
2187                       TCGReg a0, TCGReg a1, TCGReg a2)
2188{
2189    tcg_out_opc_reg(s, OPC_MULHU, a0, a1, a2);
2190}
2191
2192static const TCGOutOpBinary outop_muluh = {
2193    .base.static_constraint = C_Dynamic,
2194    .base.dynamic_constraint = cset_mulh,
2195    .out_rrr = tgen_muluh,
2196};
2197
2198static const TCGOutOpBinary outop_nand = {
2199    .base.static_constraint = C_NotImplemented,
2200};
2201
2202static const TCGOutOpBinary outop_nor = {
2203    .base.static_constraint = C_NotImplemented,
2204};
2205
2206static void tgen_or(TCGContext *s, TCGType type,
2207                    TCGReg a0, TCGReg a1, TCGReg a2)
2208{
2209    tcg_out_opc_reg(s, OPC_OR, a0, a1, a2);
2210}
2211
2212static void tgen_ori(TCGContext *s, TCGType type,
2213                     TCGReg a0, TCGReg a1, tcg_target_long a2)
2214{
2215    tcg_out_opc_imm(s, OPC_ORI, a0, a1, a2);
2216}
2217
2218static const TCGOutOpBinary outop_or = {
2219    .base.static_constraint = C_O1_I2(r, r, rI),
2220    .out_rrr = tgen_or,
2221    .out_rri = tgen_ori,
2222};
2223
2224static void tgen_orc(TCGContext *s, TCGType type,
2225                     TCGReg a0, TCGReg a1, TCGReg a2)
2226{
2227    tcg_out_opc_reg(s, OPC_ORN, a0, a1, a2);
2228}
2229
2230static const TCGOutOpBinary outop_orc = {
2231    .base.static_constraint = C_Dynamic,
2232    .base.dynamic_constraint = cset_zbb_rrr,
2233    .out_rrr = tgen_orc,
2234};
2235
2236static void tgen_rems(TCGContext *s, TCGType type,
2237                      TCGReg a0, TCGReg a1, TCGReg a2)
2238{
2239    RISCVInsn insn = type == TCG_TYPE_I32 ? OPC_REMW : OPC_REM;
2240    tcg_out_opc_reg(s, insn, a0, a1, a2);
2241}
2242
2243static const TCGOutOpBinary outop_rems = {
2244    .base.static_constraint = C_O1_I2(r, r, r),
2245    .out_rrr = tgen_rems,
2246};
2247
2248static void tgen_remu(TCGContext *s, TCGType type,
2249                      TCGReg a0, TCGReg a1, TCGReg a2)
2250{
2251    RISCVInsn insn = type == TCG_TYPE_I32 ? OPC_REMUW : OPC_REMU;
2252    tcg_out_opc_reg(s, insn, a0, a1, a2);
2253}
2254
2255static const TCGOutOpBinary outop_remu = {
2256    .base.static_constraint = C_O1_I2(r, r, r),
2257    .out_rrr = tgen_remu,
2258};
2259
2260static TCGConstraintSetIndex cset_rot(TCGType type, unsigned flags)
2261{
2262    return cpuinfo & CPUINFO_ZBB ? C_O1_I2(r, r, ri) : C_NotImplemented;
2263}
2264
2265static void tgen_rotr(TCGContext *s, TCGType type,
2266                      TCGReg a0, TCGReg a1, TCGReg a2)
2267{
2268    RISCVInsn insn = type == TCG_TYPE_I32 ? OPC_RORW : OPC_ROR;
2269    tcg_out_opc_reg(s, insn, a0, a1, a2);
2270}
2271
2272static void tgen_rotri(TCGContext *s, TCGType type,
2273                       TCGReg a0, TCGReg a1, tcg_target_long a2)
2274{
2275    RISCVInsn insn = type == TCG_TYPE_I32 ? OPC_RORIW : OPC_RORI;
2276    unsigned mask = type == TCG_TYPE_I32 ? 31 : 63;
2277    tcg_out_opc_imm(s, insn, a0, a1, a2 & mask);
2278}
2279
2280static const TCGOutOpBinary outop_rotr = {
2281    .base.static_constraint = C_Dynamic,
2282    .base.dynamic_constraint = cset_rot,
2283    .out_rrr = tgen_rotr,
2284    .out_rri = tgen_rotri,
2285};
2286
2287static void tgen_rotl(TCGContext *s, TCGType type,
2288                      TCGReg a0, TCGReg a1, TCGReg a2)
2289{
2290    RISCVInsn insn = type == TCG_TYPE_I32 ? OPC_ROLW : OPC_ROL;
2291    tcg_out_opc_reg(s, insn, a0, a1, a2);
2292}
2293
2294static void tgen_rotli(TCGContext *s, TCGType type,
2295                       TCGReg a0, TCGReg a1, tcg_target_long a2)
2296{
2297    tgen_rotri(s, type, a0, a1, -a2);
2298}
2299
2300static const TCGOutOpBinary outop_rotl = {
2301    .base.static_constraint = C_Dynamic,
2302    .base.dynamic_constraint = cset_rot,
2303    .out_rrr = tgen_rotl,
2304    .out_rri = tgen_rotli,
2305};
2306
2307static void tgen_sar(TCGContext *s, TCGType type,
2308                     TCGReg a0, TCGReg a1, TCGReg a2)
2309{
2310    RISCVInsn insn = type == TCG_TYPE_I32 ? OPC_SRAW : OPC_SRA;
2311    tcg_out_opc_reg(s, insn, a0, a1, a2);
2312}
2313
2314static void tgen_sari(TCGContext *s, TCGType type,
2315                      TCGReg a0, TCGReg a1, tcg_target_long a2)
2316{
2317    RISCVInsn insn = type == TCG_TYPE_I32 ? OPC_SRAIW : OPC_SRAI;
2318    unsigned mask = type == TCG_TYPE_I32 ? 31 : 63;
2319    tcg_out_opc_imm(s, insn, a0, a1, a2 & mask);
2320}
2321
2322static const TCGOutOpBinary outop_sar = {
2323    .base.static_constraint = C_O1_I2(r, r, ri),
2324    .out_rrr = tgen_sar,
2325    .out_rri = tgen_sari,
2326};
2327
2328static void tgen_shl(TCGContext *s, TCGType type,
2329                     TCGReg a0, TCGReg a1, TCGReg a2)
2330{
2331    RISCVInsn insn = type == TCG_TYPE_I32 ? OPC_SLLW : OPC_SLL;
2332    tcg_out_opc_reg(s, insn, a0, a1, a2);
2333}
2334
2335static void tgen_shli(TCGContext *s, TCGType type,
2336                      TCGReg a0, TCGReg a1, tcg_target_long a2)
2337{
2338    RISCVInsn insn = type == TCG_TYPE_I32 ? OPC_SLLIW : OPC_SLLI;
2339    unsigned mask = type == TCG_TYPE_I32 ? 31 : 63;
2340    tcg_out_opc_imm(s, insn, a0, a1, a2 & mask);
2341}
2342
2343static const TCGOutOpBinary outop_shl = {
2344    .base.static_constraint = C_O1_I2(r, r, ri),
2345    .out_rrr = tgen_shl,
2346    .out_rri = tgen_shli,
2347};
2348
2349static void tgen_shr(TCGContext *s, TCGType type,
2350                     TCGReg a0, TCGReg a1, TCGReg a2)
2351{
2352    RISCVInsn insn = type == TCG_TYPE_I32 ? OPC_SRLW : OPC_SRL;
2353    tcg_out_opc_reg(s, insn, a0, a1, a2);
2354}
2355
2356static void tgen_shri(TCGContext *s, TCGType type,
2357                      TCGReg a0, TCGReg a1, tcg_target_long a2)
2358{
2359    RISCVInsn insn = type == TCG_TYPE_I32 ? OPC_SRLIW : OPC_SRLI;
2360    unsigned mask = type == TCG_TYPE_I32 ? 31 : 63;
2361    tcg_out_opc_imm(s, insn, a0, a1, a2 & mask);
2362}
2363
2364static const TCGOutOpBinary outop_shr = {
2365    .base.static_constraint = C_O1_I2(r, r, ri),
2366    .out_rrr = tgen_shr,
2367    .out_rri = tgen_shri,
2368};
2369
2370static void tgen_sub(TCGContext *s, TCGType type,
2371                     TCGReg a0, TCGReg a1, TCGReg a2)
2372{
2373    RISCVInsn insn = type == TCG_TYPE_I32 ? OPC_SUBW : OPC_SUB;
2374    tcg_out_opc_reg(s, insn, a0, a1, a2);
2375}
2376
2377static const TCGOutOpSubtract outop_sub = {
2378    .base.static_constraint = C_O1_I2(r, r, r),
2379    .out_rrr = tgen_sub,
2380};
2381
2382static const TCGOutOpAddSubCarry outop_subbo = {
2383    .base.static_constraint = C_NotImplemented,
2384};
2385
2386static const TCGOutOpAddSubCarry outop_subbi = {
2387    .base.static_constraint = C_NotImplemented,
2388};
2389
2390static const TCGOutOpAddSubCarry outop_subbio = {
2391    .base.static_constraint = C_NotImplemented,
2392};
2393
2394static void tcg_out_set_borrow(TCGContext *s)
2395{
2396    g_assert_not_reached();
2397}
2398
2399static void tgen_xor(TCGContext *s, TCGType type,
2400                     TCGReg a0, TCGReg a1, TCGReg a2)
2401{
2402    tcg_out_opc_reg(s, OPC_XOR, a0, a1, a2);
2403}
2404
2405static void tgen_xori(TCGContext *s, TCGType type,
2406                      TCGReg a0, TCGReg a1, tcg_target_long a2)
2407{
2408    tcg_out_opc_imm(s, OPC_XORI, a0, a1, a2);
2409}
2410
2411static const TCGOutOpBinary outop_xor = {
2412    .base.static_constraint = C_O1_I2(r, r, rI),
2413    .out_rrr = tgen_xor,
2414    .out_rri = tgen_xori,
2415};
2416
2417static TCGConstraintSetIndex cset_bswap(TCGType type, unsigned flags)
2418{
2419    return cpuinfo & CPUINFO_ZBB ? C_O1_I1(r, r) : C_NotImplemented;
2420}
2421
2422static void tgen_bswap16(TCGContext *s, TCGType type,
2423                         TCGReg a0, TCGReg a1, unsigned flags)
2424{
2425    tcg_out_opc_imm(s, OPC_REV8, a0, a1, 0);
2426    if (flags & TCG_BSWAP_OZ) {
2427        tcg_out_opc_imm(s, OPC_SRLI, a0, a0, 48);
2428    } else {
2429        tcg_out_opc_imm(s, OPC_SRAI, a0, a0, 48);
2430    }
2431}
2432
2433static const TCGOutOpBswap outop_bswap16 = {
2434    .base.static_constraint = C_Dynamic,
2435    .base.dynamic_constraint = cset_bswap,
2436    .out_rr = tgen_bswap16,
2437};
2438
2439static void tgen_bswap32(TCGContext *s, TCGType type,
2440                         TCGReg a0, TCGReg a1, unsigned flags)
2441{
2442    tcg_out_opc_imm(s, OPC_REV8, a0, a1, 0);
2443    if (flags & TCG_BSWAP_OZ) {
2444        tcg_out_opc_imm(s, OPC_SRLI, a0, a0, 32);
2445    } else {
2446        tcg_out_opc_imm(s, OPC_SRAI, a0, a0, 32);
2447    }
2448}
2449
2450static const TCGOutOpBswap outop_bswap32 = {
2451    .base.static_constraint = C_Dynamic,
2452    .base.dynamic_constraint = cset_bswap,
2453    .out_rr = tgen_bswap32,
2454};
2455
2456static void tgen_bswap64(TCGContext *s, TCGType type, TCGReg a0, TCGReg a1)
2457{
2458    tcg_out_opc_imm(s, OPC_REV8, a0, a1, 0);
2459}
2460
2461static const TCGOutOpUnary outop_bswap64 = {
2462    .base.static_constraint = C_Dynamic,
2463    .base.dynamic_constraint = cset_bswap,
2464    .out_rr = tgen_bswap64,
2465};
2466
2467static void tgen_neg(TCGContext *s, TCGType type, TCGReg a0, TCGReg a1)
2468{
2469    tgen_sub(s, type, a0, TCG_REG_ZERO, a1);
2470}
2471
2472static const TCGOutOpUnary outop_neg = {
2473    .base.static_constraint = C_O1_I1(r, r),
2474    .out_rr = tgen_neg,
2475};
2476
2477static void tgen_not(TCGContext *s, TCGType type, TCGReg a0, TCGReg a1)
2478{
2479    tgen_xori(s, type, a0, a1, -1);
2480}
2481
2482static const TCGOutOpUnary outop_not = {
2483    .base.static_constraint = C_O1_I1(r, r),
2484    .out_rr = tgen_not,
2485};
2486
2487static const TCGOutOpDeposit outop_deposit = {
2488    .base.static_constraint = C_NotImplemented,
2489};
2490
2491static void tgen_extract(TCGContext *s, TCGType type, TCGReg a0, TCGReg a1,
2492                         unsigned ofs, unsigned len)
2493{
2494    if (ofs == 0) {
2495        switch (len) {
2496        case 16:
2497            tcg_out_ext16u(s, a0, a1);
2498            return;
2499        case 32:
2500            tcg_out_ext32u(s, a0, a1);
2501            return;
2502        }
2503    }
2504    if (ofs + len == 32) {
2505        tgen_shli(s, TCG_TYPE_I32, a0, a1, ofs);
2506        return;
2507    }
2508    if (len == 1) {
2509        tcg_out_opc_imm(s, OPC_BEXTI, a0, a1, ofs);
2510        return;
2511    }
2512    g_assert_not_reached();
2513}
2514
2515static const TCGOutOpExtract outop_extract = {
2516    .base.static_constraint = C_O1_I1(r, r),
2517    .out_rr = tgen_extract,
2518};
2519
2520static void tgen_sextract(TCGContext *s, TCGType type, TCGReg a0, TCGReg a1,
2521                          unsigned ofs, unsigned len)
2522{
2523    if (ofs == 0) {
2524        switch (len) {
2525        case 8:
2526            tcg_out_ext8s(s, type, a0, a1);
2527            return;
2528        case 16:
2529            tcg_out_ext16s(s, type, a0, a1);
2530            return;
2531        case 32:
2532            tcg_out_ext32s(s, a0, a1);
2533            return;
2534        }
2535    } else if (ofs + len == 32) {
2536        tgen_sari(s, TCG_TYPE_I32, a0, a1, ofs);
2537        return;
2538    }
2539    g_assert_not_reached();
2540}
2541
2542static const TCGOutOpExtract outop_sextract = {
2543    .base.static_constraint = C_O1_I1(r, r),
2544    .out_rr = tgen_sextract,
2545};
2546
2547static const TCGOutOpExtract2 outop_extract2 = {
2548    .base.static_constraint = C_NotImplemented,
2549};
2550
2551static void tgen_ld8u(TCGContext *s, TCGType type, TCGReg dest,
2552                      TCGReg base, ptrdiff_t offset)
2553{
2554    tcg_out_ldst(s, OPC_LBU, dest, base, offset);
2555}
2556
2557static const TCGOutOpLoad outop_ld8u = {
2558    .base.static_constraint = C_O1_I1(r, r),
2559    .out = tgen_ld8u,
2560};
2561
2562static void tgen_ld8s(TCGContext *s, TCGType type, TCGReg dest,
2563                      TCGReg base, ptrdiff_t offset)
2564{
2565    tcg_out_ldst(s, OPC_LB, dest, base, offset);
2566}
2567
2568static const TCGOutOpLoad outop_ld8s = {
2569    .base.static_constraint = C_O1_I1(r, r),
2570    .out = tgen_ld8s,
2571};
2572
2573static void tgen_ld16u(TCGContext *s, TCGType type, TCGReg dest,
2574                       TCGReg base, ptrdiff_t offset)
2575{
2576    tcg_out_ldst(s, OPC_LHU, dest, base, offset);
2577}
2578
2579static const TCGOutOpLoad outop_ld16u = {
2580    .base.static_constraint = C_O1_I1(r, r),
2581    .out = tgen_ld16u,
2582};
2583
2584static void tgen_ld16s(TCGContext *s, TCGType type, TCGReg dest,
2585                       TCGReg base, ptrdiff_t offset)
2586{
2587    tcg_out_ldst(s, OPC_LH, dest, base, offset);
2588}
2589
2590static const TCGOutOpLoad outop_ld16s = {
2591    .base.static_constraint = C_O1_I1(r, r),
2592    .out = tgen_ld16s,
2593};
2594
2595static void tgen_ld32u(TCGContext *s, TCGType type, TCGReg dest,
2596                       TCGReg base, ptrdiff_t offset)
2597{
2598    tcg_out_ldst(s, OPC_LWU, dest, base, offset);
2599}
2600
2601static const TCGOutOpLoad outop_ld32u = {
2602    .base.static_constraint = C_O1_I1(r, r),
2603    .out = tgen_ld32u,
2604};
2605
2606static void tgen_ld32s(TCGContext *s, TCGType type, TCGReg dest,
2607                       TCGReg base, ptrdiff_t offset)
2608{
2609    tcg_out_ldst(s, OPC_LW, dest, base, offset);
2610}
2611
2612static const TCGOutOpLoad outop_ld32s = {
2613    .base.static_constraint = C_O1_I1(r, r),
2614    .out = tgen_ld32s,
2615};
2616
2617static void tgen_st8_r(TCGContext *s, TCGType type, TCGReg data,
2618                       TCGReg base, ptrdiff_t offset)
2619{
2620    tcg_out_ldst(s, OPC_SB, data, base, offset);
2621}
2622
2623static const TCGOutOpStore outop_st8 = {
2624    .base.static_constraint = C_O0_I2(rz, r),
2625    .out_r = tgen_st8_r,
2626};
2627
2628static void tgen_st16_r(TCGContext *s, TCGType type, TCGReg data,
2629                        TCGReg base, ptrdiff_t offset)
2630{
2631    tcg_out_ldst(s, OPC_SH, data, base, offset);
2632}
2633
2634static const TCGOutOpStore outop_st16 = {
2635    .base.static_constraint = C_O0_I2(rz, r),
2636    .out_r = tgen_st16_r,
2637};
2638
2639static const TCGOutOpStore outop_st = {
2640    .base.static_constraint = C_O0_I2(rz, r),
2641    .out_r = tcg_out_st,
2642};
2643
2644
2645static void tcg_out_vec_op(TCGContext *s, TCGOpcode opc,
2646                           unsigned vecl, unsigned vece,
2647                           const TCGArg args[TCG_MAX_OP_ARGS],
2648                           const int const_args[TCG_MAX_OP_ARGS])
2649{
2650    TCGType type = vecl + TCG_TYPE_V64;
2651    TCGArg a0, a1, a2;
2652    int c2;
2653
2654    a0 = args[0];
2655    a1 = args[1];
2656    a2 = args[2];
2657    c2 = const_args[2];
2658
2659    switch (opc) {
2660    case INDEX_op_dupm_vec:
2661        tcg_out_dupm_vec(s, type, vece, a0, a1, a2);
2662        break;
2663    case INDEX_op_ld_vec:
2664        tcg_out_ld(s, type, a0, a1, a2);
2665        break;
2666    case INDEX_op_st_vec:
2667        tcg_out_st(s, type, a0, a1, a2);
2668        break;
2669    case INDEX_op_add_vec:
2670        set_vtype_len_sew(s, type, vece);
2671        tcg_out_opc_vv_vi(s, OPC_VADD_VV, OPC_VADD_VI, a0, a1, a2, c2);
2672        break;
2673    case INDEX_op_sub_vec:
2674        set_vtype_len_sew(s, type, vece);
2675        if (const_args[1]) {
2676            tcg_out_opc_vi(s, OPC_VRSUB_VI, a0, a2, a1);
2677        } else {
2678            tcg_out_opc_vv(s, OPC_VSUB_VV, a0, a1, a2);
2679        }
2680        break;
2681    case INDEX_op_and_vec:
2682        set_vtype_len(s, type);
2683        tcg_out_opc_vv_vi(s, OPC_VAND_VV, OPC_VAND_VI, a0, a1, a2, c2);
2684        break;
2685    case INDEX_op_or_vec:
2686        set_vtype_len(s, type);
2687        tcg_out_opc_vv_vi(s, OPC_VOR_VV, OPC_VOR_VI, a0, a1, a2, c2);
2688        break;
2689    case INDEX_op_xor_vec:
2690        set_vtype_len(s, type);
2691        tcg_out_opc_vv_vi(s, OPC_VXOR_VV, OPC_VXOR_VI, a0, a1, a2, c2);
2692        break;
2693    case INDEX_op_not_vec:
2694        set_vtype_len(s, type);
2695        tcg_out_opc_vi(s, OPC_VXOR_VI, a0, a1, -1);
2696        break;
2697    case INDEX_op_neg_vec:
2698        set_vtype_len_sew(s, type, vece);
2699        tcg_out_opc_vi(s, OPC_VRSUB_VI, a0, a1, 0);
2700        break;
2701    case INDEX_op_mul_vec:
2702        set_vtype_len_sew(s, type, vece);
2703        tcg_out_opc_vv(s, OPC_VMUL_VV, a0, a1, a2);
2704        break;
2705    case INDEX_op_ssadd_vec:
2706        set_vtype_len_sew(s, type, vece);
2707        tcg_out_opc_vv_vi(s, OPC_VSADD_VV, OPC_VSADD_VI, a0, a1, a2, c2);
2708        break;
2709    case INDEX_op_sssub_vec:
2710        set_vtype_len_sew(s, type, vece);
2711        tcg_out_opc_vv_vi(s, OPC_VSSUB_VV, OPC_VSSUB_VI, a0, a1, a2, c2);
2712        break;
2713    case INDEX_op_usadd_vec:
2714        set_vtype_len_sew(s, type, vece);
2715        tcg_out_opc_vv_vi(s, OPC_VSADDU_VV, OPC_VSADDU_VI, a0, a1, a2, c2);
2716        break;
2717    case INDEX_op_ussub_vec:
2718        set_vtype_len_sew(s, type, vece);
2719        tcg_out_opc_vv_vi(s, OPC_VSSUBU_VV, OPC_VSSUBU_VI, a0, a1, a2, c2);
2720        break;
2721    case INDEX_op_smax_vec:
2722        set_vtype_len_sew(s, type, vece);
2723        tcg_out_opc_vv_vi(s, OPC_VMAX_VV, OPC_VMAX_VI, a0, a1, a2, c2);
2724        break;
2725    case INDEX_op_smin_vec:
2726        set_vtype_len_sew(s, type, vece);
2727        tcg_out_opc_vv_vi(s, OPC_VMIN_VV, OPC_VMIN_VI, a0, a1, a2, c2);
2728        break;
2729    case INDEX_op_umax_vec:
2730        set_vtype_len_sew(s, type, vece);
2731        tcg_out_opc_vv_vi(s, OPC_VMAXU_VV, OPC_VMAXU_VI, a0, a1, a2, c2);
2732        break;
2733    case INDEX_op_umin_vec:
2734        set_vtype_len_sew(s, type, vece);
2735        tcg_out_opc_vv_vi(s, OPC_VMINU_VV, OPC_VMINU_VI, a0, a1, a2, c2);
2736        break;
2737    case INDEX_op_shls_vec:
2738        set_vtype_len_sew(s, type, vece);
2739        tcg_out_opc_vx(s, OPC_VSLL_VX, a0, a1, a2);
2740        break;
2741    case INDEX_op_shrs_vec:
2742        set_vtype_len_sew(s, type, vece);
2743        tcg_out_opc_vx(s, OPC_VSRL_VX, a0, a1, a2);
2744        break;
2745    case INDEX_op_sars_vec:
2746        set_vtype_len_sew(s, type, vece);
2747        tcg_out_opc_vx(s, OPC_VSRA_VX, a0, a1, a2);
2748        break;
2749    case INDEX_op_shlv_vec:
2750        set_vtype_len_sew(s, type, vece);
2751        tcg_out_opc_vv(s, OPC_VSLL_VV, a0, a1, a2);
2752        break;
2753    case INDEX_op_shrv_vec:
2754        set_vtype_len_sew(s, type, vece);
2755        tcg_out_opc_vv(s, OPC_VSRL_VV, a0, a1, a2);
2756        break;
2757    case INDEX_op_sarv_vec:
2758        set_vtype_len_sew(s, type, vece);
2759        tcg_out_opc_vv(s, OPC_VSRA_VV, a0, a1, a2);
2760        break;
2761    case INDEX_op_shli_vec:
2762        set_vtype_len_sew(s, type, vece);
2763        tcg_out_vshifti(s, OPC_VSLL_VI, OPC_VSLL_VX, a0, a1, a2);
2764        break;
2765    case INDEX_op_shri_vec:
2766        set_vtype_len_sew(s, type, vece);
2767        tcg_out_vshifti(s, OPC_VSRL_VI, OPC_VSRL_VX, a0, a1, a2);
2768        break;
2769    case INDEX_op_sari_vec:
2770        set_vtype_len_sew(s, type, vece);
2771        tcg_out_vshifti(s, OPC_VSRA_VI, OPC_VSRA_VX, a0, a1, a2);
2772        break;
2773    case INDEX_op_rotli_vec:
2774        set_vtype_len_sew(s, type, vece);
2775        tcg_out_vshifti(s, OPC_VSLL_VI, OPC_VSLL_VX, TCG_REG_V0, a1, a2);
2776        tcg_out_vshifti(s, OPC_VSRL_VI, OPC_VSRL_VX, a0, a1,
2777                        -a2 & ((8 << vece) - 1));
2778        tcg_out_opc_vv(s, OPC_VOR_VV, a0, a0, TCG_REG_V0);
2779        break;
2780    case INDEX_op_rotls_vec:
2781        set_vtype_len_sew(s, type, vece);
2782        tcg_out_opc_vx(s, OPC_VSLL_VX, TCG_REG_V0, a1, a2);
2783        tcg_out_opc_reg(s, OPC_SUBW, TCG_REG_TMP0, TCG_REG_ZERO, a2);
2784        tcg_out_opc_vx(s, OPC_VSRL_VX, a0, a1, TCG_REG_TMP0);
2785        tcg_out_opc_vv(s, OPC_VOR_VV, a0, a0, TCG_REG_V0);
2786        break;
2787    case INDEX_op_rotlv_vec:
2788        set_vtype_len_sew(s, type, vece);
2789        tcg_out_opc_vi(s, OPC_VRSUB_VI, TCG_REG_V0, a2, 0);
2790        tcg_out_opc_vv(s, OPC_VSRL_VV, TCG_REG_V0, a1, TCG_REG_V0);
2791        tcg_out_opc_vv(s, OPC_VSLL_VV, a0, a1, a2);
2792        tcg_out_opc_vv(s, OPC_VOR_VV, a0, a0, TCG_REG_V0);
2793        break;
2794    case INDEX_op_rotrv_vec:
2795        set_vtype_len_sew(s, type, vece);
2796        tcg_out_opc_vi(s, OPC_VRSUB_VI, TCG_REG_V0, a2, 0);
2797        tcg_out_opc_vv(s, OPC_VSLL_VV, TCG_REG_V0, a1, TCG_REG_V0);
2798        tcg_out_opc_vv(s, OPC_VSRL_VV, a0, a1, a2);
2799        tcg_out_opc_vv(s, OPC_VOR_VV, a0, a0, TCG_REG_V0);
2800        break;
2801    case INDEX_op_cmp_vec:
2802        tcg_out_cmpsel(s, type, vece, args[3], a0, a1, a2, c2,
2803                       -1, true, 0, true);
2804        break;
2805    case INDEX_op_cmpsel_vec:
2806        tcg_out_cmpsel(s, type, vece, args[5], a0, a1, a2, c2,
2807                       args[3], const_args[3], args[4], const_args[4]);
2808        break;
2809    case INDEX_op_mov_vec: /* Always emitted via tcg_out_mov.  */
2810    case INDEX_op_dup_vec: /* Always emitted via tcg_out_dup_vec.  */
2811    default:
2812        g_assert_not_reached();
2813    }
2814}
2815
2816void tcg_expand_vec_op(TCGOpcode opc, TCGType type, unsigned vece,
2817                       TCGArg a0, ...)
2818{
2819    g_assert_not_reached();
2820}
2821
2822int tcg_can_emit_vec_op(TCGOpcode opc, TCGType type, unsigned vece)
2823{
2824    switch (opc) {
2825    case INDEX_op_add_vec:
2826    case INDEX_op_sub_vec:
2827    case INDEX_op_and_vec:
2828    case INDEX_op_or_vec:
2829    case INDEX_op_xor_vec:
2830    case INDEX_op_not_vec:
2831    case INDEX_op_neg_vec:
2832    case INDEX_op_mul_vec:
2833    case INDEX_op_ssadd_vec:
2834    case INDEX_op_sssub_vec:
2835    case INDEX_op_usadd_vec:
2836    case INDEX_op_ussub_vec:
2837    case INDEX_op_smax_vec:
2838    case INDEX_op_smin_vec:
2839    case INDEX_op_umax_vec:
2840    case INDEX_op_umin_vec:
2841    case INDEX_op_shls_vec:
2842    case INDEX_op_shrs_vec:
2843    case INDEX_op_sars_vec:
2844    case INDEX_op_shlv_vec:
2845    case INDEX_op_shrv_vec:
2846    case INDEX_op_sarv_vec:
2847    case INDEX_op_shri_vec:
2848    case INDEX_op_shli_vec:
2849    case INDEX_op_sari_vec:
2850    case INDEX_op_rotls_vec:
2851    case INDEX_op_rotlv_vec:
2852    case INDEX_op_rotrv_vec:
2853    case INDEX_op_rotli_vec:
2854    case INDEX_op_cmp_vec:
2855    case INDEX_op_cmpsel_vec:
2856        return 1;
2857    default:
2858        return 0;
2859    }
2860}
2861
2862static TCGConstraintSetIndex
2863tcg_target_op_def(TCGOpcode op, TCGType type, unsigned flags)
2864{
2865    switch (op) {
2866    case INDEX_op_st_vec:
2867        return C_O0_I2(v, r);
2868    case INDEX_op_dup_vec:
2869    case INDEX_op_dupm_vec:
2870    case INDEX_op_ld_vec:
2871        return C_O1_I1(v, r);
2872    case INDEX_op_neg_vec:
2873    case INDEX_op_not_vec:
2874    case INDEX_op_shli_vec:
2875    case INDEX_op_shri_vec:
2876    case INDEX_op_sari_vec:
2877    case INDEX_op_rotli_vec:
2878        return C_O1_I1(v, v);
2879    case INDEX_op_add_vec:
2880    case INDEX_op_and_vec:
2881    case INDEX_op_or_vec:
2882    case INDEX_op_xor_vec:
2883    case INDEX_op_ssadd_vec:
2884    case INDEX_op_sssub_vec:
2885    case INDEX_op_usadd_vec:
2886    case INDEX_op_ussub_vec:
2887    case INDEX_op_smax_vec:
2888    case INDEX_op_smin_vec:
2889    case INDEX_op_umax_vec:
2890    case INDEX_op_umin_vec:
2891        return C_O1_I2(v, v, vK);
2892    case INDEX_op_sub_vec:
2893        return C_O1_I2(v, vK, v);
2894    case INDEX_op_mul_vec:
2895    case INDEX_op_shlv_vec:
2896    case INDEX_op_shrv_vec:
2897    case INDEX_op_sarv_vec:
2898    case INDEX_op_rotlv_vec:
2899    case INDEX_op_rotrv_vec:
2900        return C_O1_I2(v, v, v);
2901    case INDEX_op_shls_vec:
2902    case INDEX_op_shrs_vec:
2903    case INDEX_op_sars_vec:
2904    case INDEX_op_rotls_vec:
2905        return C_O1_I2(v, v, r);
2906    case INDEX_op_cmp_vec:
2907        return C_O1_I2(v, v, vL);
2908    case INDEX_op_cmpsel_vec:
2909        return C_O1_I4(v, v, vL, vK, vK);
2910    default:
2911        return C_NotImplemented;
2912    }
2913}
2914
2915static const int tcg_target_callee_save_regs[] = {
2916    TCG_REG_S0,       /* used for the global env (TCG_AREG0) */
2917    TCG_REG_S1,
2918    TCG_REG_S2,
2919    TCG_REG_S3,
2920    TCG_REG_S4,
2921    TCG_REG_S5,
2922    TCG_REG_S6,
2923    TCG_REG_S7,
2924    TCG_REG_S8,
2925    TCG_REG_S9,
2926    TCG_REG_S10,
2927    TCG_REG_S11,
2928    TCG_REG_RA,       /* should be last for ABI compliance */
2929};
2930
2931/* Stack frame parameters.  */
2932#define REG_SIZE   (TCG_TARGET_REG_BITS / 8)
2933#define SAVE_SIZE  ((int)ARRAY_SIZE(tcg_target_callee_save_regs) * REG_SIZE)
2934#define TEMP_SIZE  (CPU_TEMP_BUF_NLONGS * (int)sizeof(long))
2935#define FRAME_SIZE ((TCG_STATIC_CALL_ARGS_SIZE + TEMP_SIZE + SAVE_SIZE \
2936                     + TCG_TARGET_STACK_ALIGN - 1) \
2937                    & -TCG_TARGET_STACK_ALIGN)
2938#define SAVE_OFS   (TCG_STATIC_CALL_ARGS_SIZE + TEMP_SIZE)
2939
2940/* We're expecting to be able to use an immediate for frame allocation.  */
2941QEMU_BUILD_BUG_ON(FRAME_SIZE > 0x7ff);
2942
2943/* Generate global QEMU prologue and epilogue code */
2944static void tcg_target_qemu_prologue(TCGContext *s)
2945{
2946    int i;
2947
2948    tcg_set_frame(s, TCG_REG_SP, TCG_STATIC_CALL_ARGS_SIZE, TEMP_SIZE);
2949
2950    /* TB prologue */
2951    tcg_out_opc_imm(s, OPC_ADDI, TCG_REG_SP, TCG_REG_SP, -FRAME_SIZE);
2952    for (i = 0; i < ARRAY_SIZE(tcg_target_callee_save_regs); i++) {
2953        tcg_out_st(s, TCG_TYPE_REG, tcg_target_callee_save_regs[i],
2954                   TCG_REG_SP, SAVE_OFS + i * REG_SIZE);
2955    }
2956
2957    if (!tcg_use_softmmu && guest_base) {
2958        tcg_out_movi(s, TCG_TYPE_PTR, TCG_GUEST_BASE_REG, guest_base);
2959        tcg_regset_set_reg(s->reserved_regs, TCG_GUEST_BASE_REG);
2960    }
2961
2962    /* Call generated code */
2963    tcg_out_mov(s, TCG_TYPE_PTR, TCG_AREG0, tcg_target_call_iarg_regs[0]);
2964    tcg_out_opc_imm(s, OPC_JALR, TCG_REG_ZERO, tcg_target_call_iarg_regs[1], 0);
2965
2966    /* Return path for goto_ptr. Set return value to 0 */
2967    tcg_code_gen_epilogue = tcg_splitwx_to_rx(s->code_ptr);
2968    tcg_out_mov(s, TCG_TYPE_REG, TCG_REG_A0, TCG_REG_ZERO);
2969
2970    /* TB epilogue */
2971    tb_ret_addr = tcg_splitwx_to_rx(s->code_ptr);
2972    for (i = 0; i < ARRAY_SIZE(tcg_target_callee_save_regs); i++) {
2973        tcg_out_ld(s, TCG_TYPE_REG, tcg_target_callee_save_regs[i],
2974                   TCG_REG_SP, SAVE_OFS + i * REG_SIZE);
2975    }
2976
2977    tcg_out_opc_imm(s, OPC_ADDI, TCG_REG_SP, TCG_REG_SP, FRAME_SIZE);
2978    tcg_out_opc_imm(s, OPC_JALR, TCG_REG_ZERO, TCG_REG_RA, 0);
2979}
2980
2981static void tcg_out_tb_start(TCGContext *s)
2982{
2983    init_setting_vtype(s);
2984}
2985
2986static bool vtype_check(unsigned vtype)
2987{
2988    unsigned long tmp;
2989
2990    /* vsetvl tmp, zero, vtype */
2991    asm(".insn r 0x57, 7, 0x40, %0, zero, %1" : "=r"(tmp) : "r"(vtype));
2992    return tmp != 0;
2993}
2994
2995static void probe_frac_lmul_1(TCGType type, MemOp vsew)
2996{
2997    VsetCache *p = &riscv_vset_cache[type - TCG_TYPE_V64][vsew];
2998    unsigned avl = tcg_type_size(type) >> vsew;
2999    int lmul = type - riscv_lg2_vlenb;
3000    unsigned vtype = encode_vtype(true, true, vsew, lmul & 7);
3001    bool lmul_eq_avl = true;
3002
3003    /* Guaranteed by Zve64x. */
3004    assert(lmul < 3);
3005
3006    /*
3007     * For LMUL < -3, the host vector size is so large that TYPE
3008     * is smaller than the minimum 1/8 fraction.
3009     *
3010     * For other fractional LMUL settings, implementations must
3011     * support SEW settings between SEW_MIN and LMUL * ELEN, inclusive.
3012     * So if ELEN = 64, LMUL = 1/2, then SEW will support e8, e16, e32,
3013     * but e64 may not be supported. In other words, the hardware only
3014     * guarantees SEW_MIN <= SEW <= LMUL * ELEN.  Check.
3015     */
3016    if (lmul < 0 && (lmul < -3 || !vtype_check(vtype))) {
3017        vtype = encode_vtype(true, true, vsew, VLMUL_M1);
3018        lmul_eq_avl = false;
3019    }
3020
3021    if (avl < 32) {
3022        p->vset_insn = encode_vseti(OPC_VSETIVLI, TCG_REG_ZERO, avl, vtype);
3023    } else if (lmul_eq_avl) {
3024        /* rd != 0 and rs1 == 0 uses vlmax */
3025        p->vset_insn = encode_vset(OPC_VSETVLI, TCG_REG_TMP0, TCG_REG_ZERO, vtype);
3026    } else {
3027        p->movi_insn = encode_i(OPC_ADDI, TCG_REG_TMP0, TCG_REG_ZERO, avl);
3028        p->vset_insn = encode_vset(OPC_VSETVLI, TCG_REG_ZERO, TCG_REG_TMP0, vtype);
3029    }
3030}
3031
3032static void probe_frac_lmul(void)
3033{
3034    /* Match riscv_lg2_vlenb to TCG_TYPE_V64. */
3035    QEMU_BUILD_BUG_ON(TCG_TYPE_V64 != 3);
3036
3037    for (TCGType t = TCG_TYPE_V64; t <= TCG_TYPE_V256; t++) {
3038        for (MemOp e = MO_8; e <= MO_64; e++) {
3039            probe_frac_lmul_1(t, e);
3040        }
3041    }
3042}
3043
3044static void tcg_target_init(TCGContext *s)
3045{
3046    tcg_target_available_regs[TCG_TYPE_I32] = 0xffffffff;
3047    tcg_target_available_regs[TCG_TYPE_I64] = 0xffffffff;
3048
3049    tcg_target_call_clobber_regs = -1;
3050    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_S0);
3051    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_S1);
3052    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_S2);
3053    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_S3);
3054    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_S4);
3055    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_S5);
3056    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_S6);
3057    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_S7);
3058    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_S8);
3059    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_S9);
3060    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_S10);
3061    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_S11);
3062
3063    s->reserved_regs = 0;
3064    tcg_regset_set_reg(s->reserved_regs, TCG_REG_ZERO);
3065    tcg_regset_set_reg(s->reserved_regs, TCG_REG_TMP0);
3066    tcg_regset_set_reg(s->reserved_regs, TCG_REG_TMP1);
3067    tcg_regset_set_reg(s->reserved_regs, TCG_REG_TMP2);
3068    tcg_regset_set_reg(s->reserved_regs, TCG_REG_SP);
3069    tcg_regset_set_reg(s->reserved_regs, TCG_REG_GP);
3070    tcg_regset_set_reg(s->reserved_regs, TCG_REG_TP);
3071
3072    if (cpuinfo & CPUINFO_ZVE64X) {
3073        switch (riscv_lg2_vlenb) {
3074        case TCG_TYPE_V64:
3075            tcg_target_available_regs[TCG_TYPE_V64] = ALL_VECTOR_REGS;
3076            tcg_target_available_regs[TCG_TYPE_V128] = ALL_DVECTOR_REG_GROUPS;
3077            tcg_target_available_regs[TCG_TYPE_V256] = ALL_QVECTOR_REG_GROUPS;
3078            s->reserved_regs |= (~ALL_QVECTOR_REG_GROUPS & ALL_VECTOR_REGS);
3079            break;
3080        case TCG_TYPE_V128:
3081            tcg_target_available_regs[TCG_TYPE_V64] = ALL_VECTOR_REGS;
3082            tcg_target_available_regs[TCG_TYPE_V128] = ALL_VECTOR_REGS;
3083            tcg_target_available_regs[TCG_TYPE_V256] = ALL_DVECTOR_REG_GROUPS;
3084            s->reserved_regs |= (~ALL_DVECTOR_REG_GROUPS & ALL_VECTOR_REGS);
3085            break;
3086        default:
3087            /* Guaranteed by Zve64x. */
3088            tcg_debug_assert(riscv_lg2_vlenb >= TCG_TYPE_V256);
3089            tcg_target_available_regs[TCG_TYPE_V64] = ALL_VECTOR_REGS;
3090            tcg_target_available_regs[TCG_TYPE_V128] = ALL_VECTOR_REGS;
3091            tcg_target_available_regs[TCG_TYPE_V256] = ALL_VECTOR_REGS;
3092            break;
3093        }
3094        tcg_regset_set_reg(s->reserved_regs, TCG_REG_V0);
3095        probe_frac_lmul();
3096    }
3097}
3098
3099typedef struct {
3100    DebugFrameHeader h;
3101    uint8_t fde_def_cfa[4];
3102    uint8_t fde_reg_ofs[ARRAY_SIZE(tcg_target_callee_save_regs) * 2];
3103} DebugFrame;
3104
3105#define ELF_HOST_MACHINE EM_RISCV
3106
3107static const DebugFrame debug_frame = {
3108    .h.cie.len = sizeof(DebugFrameCIE) - 4, /* length after .len member */
3109    .h.cie.id = -1,
3110    .h.cie.version = 1,
3111    .h.cie.code_align = 1,
3112    .h.cie.data_align = -(TCG_TARGET_REG_BITS / 8) & 0x7f, /* sleb128 */
3113    .h.cie.return_column = TCG_REG_RA,
3114
3115    /* Total FDE size does not include the "len" member.  */
3116    .h.fde.len = sizeof(DebugFrame) - offsetof(DebugFrame, h.fde.cie_offset),
3117
3118    .fde_def_cfa = {
3119        12, TCG_REG_SP,                 /* DW_CFA_def_cfa sp, ... */
3120        (FRAME_SIZE & 0x7f) | 0x80,     /* ... uleb128 FRAME_SIZE */
3121        (FRAME_SIZE >> 7)
3122    },
3123    .fde_reg_ofs = {
3124        0x80 + 9,  12,                  /* DW_CFA_offset, s1,  -96 */
3125        0x80 + 18, 11,                  /* DW_CFA_offset, s2,  -88 */
3126        0x80 + 19, 10,                  /* DW_CFA_offset, s3,  -80 */
3127        0x80 + 20, 9,                   /* DW_CFA_offset, s4,  -72 */
3128        0x80 + 21, 8,                   /* DW_CFA_offset, s5,  -64 */
3129        0x80 + 22, 7,                   /* DW_CFA_offset, s6,  -56 */
3130        0x80 + 23, 6,                   /* DW_CFA_offset, s7,  -48 */
3131        0x80 + 24, 5,                   /* DW_CFA_offset, s8,  -40 */
3132        0x80 + 25, 4,                   /* DW_CFA_offset, s9,  -32 */
3133        0x80 + 26, 3,                   /* DW_CFA_offset, s10, -24 */
3134        0x80 + 27, 2,                   /* DW_CFA_offset, s11, -16 */
3135        0x80 + 1 , 1,                   /* DW_CFA_offset, ra,  -8 */
3136    }
3137};
3138
3139void tcg_register_jit(const void *buf, size_t buf_size)
3140{
3141    tcg_register_jit_int(buf, buf_size, &debug_frame, sizeof(debug_frame));
3142}
3143