xref: /openbmc/qemu/tcg/riscv/tcg-target.c.inc (revision 76f42780292c16a0d2f36cbbfbaf57495cd4d5e8)
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 const struct {
1111    RISCVInsn op;
1112    bool swap;
1113} tcg_brcond_to_riscv[] = {
1114    [TCG_COND_EQ] =  { OPC_BEQ,  false },
1115    [TCG_COND_NE] =  { OPC_BNE,  false },
1116    [TCG_COND_LT] =  { OPC_BLT,  false },
1117    [TCG_COND_GE] =  { OPC_BGE,  false },
1118    [TCG_COND_LE] =  { OPC_BGE,  true  },
1119    [TCG_COND_GT] =  { OPC_BLT,  true  },
1120    [TCG_COND_LTU] = { OPC_BLTU, false },
1121    [TCG_COND_GEU] = { OPC_BGEU, false },
1122    [TCG_COND_LEU] = { OPC_BGEU, true  },
1123    [TCG_COND_GTU] = { OPC_BLTU, true  }
1124};
1125
1126static void tgen_brcond(TCGContext *s, TCGType type, TCGCond cond,
1127                        TCGReg arg1, TCGReg arg2, TCGLabel *l)
1128{
1129    RISCVInsn op = tcg_brcond_to_riscv[cond].op;
1130
1131    tcg_debug_assert(op != 0);
1132
1133    if (tcg_brcond_to_riscv[cond].swap) {
1134        TCGReg t = arg1;
1135        arg1 = arg2;
1136        arg2 = t;
1137    }
1138
1139    tcg_out_reloc(s, s->code_ptr, R_RISCV_BRANCH, l, 0);
1140    tcg_out_opc_branch(s, op, arg1, arg2, 0);
1141}
1142
1143static const TCGOutOpBrcond outop_brcond = {
1144    .base.static_constraint = C_O0_I2(r, rz),
1145    .out_rr = tgen_brcond,
1146};
1147
1148#define SETCOND_INV    TCG_TARGET_NB_REGS
1149#define SETCOND_NEZ    (SETCOND_INV << 1)
1150#define SETCOND_FLAGS  (SETCOND_INV | SETCOND_NEZ)
1151
1152static int tcg_out_setcond_int(TCGContext *s, TCGCond cond, TCGReg ret,
1153                               TCGReg arg1, tcg_target_long arg2, bool c2)
1154{
1155    int flags = 0;
1156
1157    switch (cond) {
1158    case TCG_COND_EQ:    /* -> NE  */
1159    case TCG_COND_GE:    /* -> LT  */
1160    case TCG_COND_GEU:   /* -> LTU */
1161    case TCG_COND_GT:    /* -> LE  */
1162    case TCG_COND_GTU:   /* -> LEU */
1163        cond = tcg_invert_cond(cond);
1164        flags ^= SETCOND_INV;
1165        break;
1166    default:
1167        break;
1168    }
1169
1170    switch (cond) {
1171    case TCG_COND_LE:
1172    case TCG_COND_LEU:
1173        /*
1174         * If we have a constant input, the most efficient way to implement
1175         * LE is by adding 1 and using LT.  Watch out for wrap around for LEU.
1176         * We don't need to care for this for LE because the constant input
1177         * is constrained to signed 12-bit, and 0x800 is representable in the
1178         * temporary register.
1179         */
1180        if (c2) {
1181            if (cond == TCG_COND_LEU) {
1182                /* unsigned <= -1 is true */
1183                if (arg2 == -1) {
1184                    tcg_out_movi(s, TCG_TYPE_REG, ret, !(flags & SETCOND_INV));
1185                    return ret;
1186                }
1187                cond = TCG_COND_LTU;
1188            } else {
1189                cond = TCG_COND_LT;
1190            }
1191            tcg_debug_assert(arg2 <= 0x7ff);
1192            if (++arg2 == 0x800) {
1193                tcg_out_movi(s, TCG_TYPE_REG, TCG_REG_TMP0, arg2);
1194                arg2 = TCG_REG_TMP0;
1195                c2 = false;
1196            }
1197        } else {
1198            TCGReg tmp = arg2;
1199            arg2 = arg1;
1200            arg1 = tmp;
1201            cond = tcg_swap_cond(cond);    /* LE -> GE */
1202            cond = tcg_invert_cond(cond);  /* GE -> LT */
1203            flags ^= SETCOND_INV;
1204        }
1205        break;
1206    default:
1207        break;
1208    }
1209
1210    switch (cond) {
1211    case TCG_COND_NE:
1212        flags |= SETCOND_NEZ;
1213        if (!c2) {
1214            tcg_out_opc_reg(s, OPC_XOR, ret, arg1, arg2);
1215        } else if (arg2 == 0) {
1216            ret = arg1;
1217        } else {
1218            tcg_out_opc_imm(s, OPC_XORI, ret, arg1, arg2);
1219        }
1220        break;
1221
1222    case TCG_COND_LT:
1223        if (c2) {
1224            tcg_out_opc_imm(s, OPC_SLTI, ret, arg1, arg2);
1225        } else {
1226            tcg_out_opc_reg(s, OPC_SLT, ret, arg1, arg2);
1227        }
1228        break;
1229
1230    case TCG_COND_LTU:
1231        if (c2) {
1232            tcg_out_opc_imm(s, OPC_SLTIU, ret, arg1, arg2);
1233        } else {
1234            tcg_out_opc_reg(s, OPC_SLTU, ret, arg1, arg2);
1235        }
1236        break;
1237
1238    default:
1239        g_assert_not_reached();
1240    }
1241
1242    return ret | flags;
1243}
1244
1245static void tcg_out_setcond(TCGContext *s, TCGCond cond, TCGReg ret,
1246                            TCGReg arg1, tcg_target_long arg2, bool c2)
1247{
1248    int tmpflags = tcg_out_setcond_int(s, cond, ret, arg1, arg2, c2);
1249
1250    if (tmpflags != ret) {
1251        TCGReg tmp = tmpflags & ~SETCOND_FLAGS;
1252
1253        switch (tmpflags & SETCOND_FLAGS) {
1254        case SETCOND_INV:
1255            /* Intermediate result is boolean: simply invert. */
1256            tcg_out_opc_imm(s, OPC_XORI, ret, tmp, 1);
1257            break;
1258        case SETCOND_NEZ:
1259            /* Intermediate result is zero/non-zero: test != 0. */
1260            tcg_out_opc_reg(s, OPC_SLTU, ret, TCG_REG_ZERO, tmp);
1261            break;
1262        case SETCOND_NEZ | SETCOND_INV:
1263            /* Intermediate result is zero/non-zero: test == 0. */
1264            tcg_out_opc_imm(s, OPC_SLTIU, ret, tmp, 1);
1265            break;
1266        default:
1267            g_assert_not_reached();
1268        }
1269    }
1270}
1271
1272static void tgen_setcond(TCGContext *s, TCGType type, TCGCond cond,
1273                         TCGReg dest, TCGReg arg1, TCGReg arg2)
1274{
1275    tcg_out_setcond(s, cond, dest, arg1, arg2, false);
1276}
1277
1278static void tgen_setcondi(TCGContext *s, TCGType type, TCGCond cond,
1279                          TCGReg dest, TCGReg arg1, tcg_target_long arg2)
1280{
1281    tcg_out_setcond(s, cond, dest, arg1, arg2, true);
1282}
1283
1284static const TCGOutOpSetcond outop_setcond = {
1285    .base.static_constraint = C_O1_I2(r, r, rI),
1286    .out_rrr = tgen_setcond,
1287    .out_rri = tgen_setcondi,
1288};
1289
1290static void tcg_out_negsetcond(TCGContext *s, TCGCond cond, TCGReg ret,
1291                               TCGReg arg1, tcg_target_long arg2, bool c2)
1292{
1293    int tmpflags;
1294    TCGReg tmp;
1295
1296    /* For LT/GE comparison against 0, replicate the sign bit. */
1297    if (c2 && arg2 == 0) {
1298        switch (cond) {
1299        case TCG_COND_GE:
1300            tcg_out_opc_imm(s, OPC_XORI, ret, arg1, -1);
1301            arg1 = ret;
1302            /* fall through */
1303        case TCG_COND_LT:
1304            tcg_out_opc_imm(s, OPC_SRAI, ret, arg1, TCG_TARGET_REG_BITS - 1);
1305            return;
1306        default:
1307            break;
1308        }
1309    }
1310
1311    tmpflags = tcg_out_setcond_int(s, cond, ret, arg1, arg2, c2);
1312    tmp = tmpflags & ~SETCOND_FLAGS;
1313
1314    /* If intermediate result is zero/non-zero: test != 0. */
1315    if (tmpflags & SETCOND_NEZ) {
1316        tcg_out_opc_reg(s, OPC_SLTU, ret, TCG_REG_ZERO, tmp);
1317        tmp = ret;
1318    }
1319
1320    /* Produce the 0/-1 result. */
1321    if (tmpflags & SETCOND_INV) {
1322        tcg_out_opc_imm(s, OPC_ADDI, ret, tmp, -1);
1323    } else {
1324        tcg_out_opc_reg(s, OPC_SUB, ret, TCG_REG_ZERO, tmp);
1325    }
1326}
1327
1328static void tgen_negsetcond(TCGContext *s, TCGType type, TCGCond cond,
1329                            TCGReg dest, TCGReg arg1, TCGReg arg2)
1330{
1331    tcg_out_negsetcond(s, cond, dest, arg1, arg2, false);
1332}
1333
1334static void tgen_negsetcondi(TCGContext *s, TCGType type, TCGCond cond,
1335                             TCGReg dest, TCGReg arg1, tcg_target_long arg2)
1336{
1337    tcg_out_negsetcond(s, cond, dest, arg1, arg2, true);
1338}
1339
1340static const TCGOutOpSetcond outop_negsetcond = {
1341    .base.static_constraint = C_O1_I2(r, r, rI),
1342    .out_rrr = tgen_negsetcond,
1343    .out_rri = tgen_negsetcondi,
1344};
1345
1346static void tcg_out_movcond_zicond(TCGContext *s, TCGReg ret, TCGReg test_ne,
1347                                   int val1, bool c_val1,
1348                                   int val2, bool c_val2)
1349{
1350    if (val1 == 0) {
1351        if (c_val2) {
1352            tcg_out_movi(s, TCG_TYPE_REG, TCG_REG_TMP1, val2);
1353            val2 = TCG_REG_TMP1;
1354        }
1355        tcg_out_opc_reg(s, OPC_CZERO_NEZ, ret, val2, test_ne);
1356        return;
1357    }
1358
1359    if (val2 == 0) {
1360        if (c_val1) {
1361            tcg_out_movi(s, TCG_TYPE_REG, TCG_REG_TMP1, val1);
1362            val1 = TCG_REG_TMP1;
1363        }
1364        tcg_out_opc_reg(s, OPC_CZERO_EQZ, ret, val1, test_ne);
1365        return;
1366    }
1367
1368    if (c_val2) {
1369        if (c_val1) {
1370            tcg_out_movi(s, TCG_TYPE_REG, TCG_REG_TMP1, val1 - val2);
1371        } else {
1372            tcg_out_opc_imm(s, OPC_ADDI, TCG_REG_TMP1, val1, -val2);
1373        }
1374        tcg_out_opc_reg(s, OPC_CZERO_EQZ, ret, TCG_REG_TMP1, test_ne);
1375        tcg_out_opc_imm(s, OPC_ADDI, ret, ret, val2);
1376        return;
1377    }
1378
1379    if (c_val1) {
1380        tcg_out_opc_imm(s, OPC_ADDI, TCG_REG_TMP1, val2, -val1);
1381        tcg_out_opc_reg(s, OPC_CZERO_NEZ, ret, TCG_REG_TMP1, test_ne);
1382        tcg_out_opc_imm(s, OPC_ADDI, ret, ret, val1);
1383        return;
1384    }
1385
1386    tcg_out_opc_reg(s, OPC_CZERO_NEZ, TCG_REG_TMP1, val2, test_ne);
1387    tcg_out_opc_reg(s, OPC_CZERO_EQZ, TCG_REG_TMP0, val1, test_ne);
1388    tcg_out_opc_reg(s, OPC_OR, ret, TCG_REG_TMP0, TCG_REG_TMP1);
1389}
1390
1391static void tcg_out_movcond_br1(TCGContext *s, TCGCond cond, TCGReg ret,
1392                                TCGReg cmp1, TCGReg cmp2,
1393                                int val, bool c_val)
1394{
1395    RISCVInsn op;
1396    int disp = 8;
1397
1398    tcg_debug_assert((unsigned)cond < ARRAY_SIZE(tcg_brcond_to_riscv));
1399    op = tcg_brcond_to_riscv[cond].op;
1400    tcg_debug_assert(op != 0);
1401
1402    if (tcg_brcond_to_riscv[cond].swap) {
1403        tcg_out_opc_branch(s, op, cmp2, cmp1, disp);
1404    } else {
1405        tcg_out_opc_branch(s, op, cmp1, cmp2, disp);
1406    }
1407    if (c_val) {
1408        tcg_out_opc_imm(s, OPC_ADDI, ret, TCG_REG_ZERO, val);
1409    } else {
1410        tcg_out_opc_imm(s, OPC_ADDI, ret, val, 0);
1411    }
1412}
1413
1414static void tcg_out_movcond_br2(TCGContext *s, TCGCond cond, TCGReg ret,
1415                                TCGReg cmp1, TCGReg cmp2,
1416                                int val1, bool c_val1,
1417                                int val2, bool c_val2)
1418{
1419    TCGReg tmp;
1420
1421    /* TCG optimizer reorders to prefer ret matching val2. */
1422    if (!c_val2 && ret == val2) {
1423        cond = tcg_invert_cond(cond);
1424        tcg_out_movcond_br1(s, cond, ret, cmp1, cmp2, val1, c_val1);
1425        return;
1426    }
1427
1428    if (!c_val1 && ret == val1) {
1429        tcg_out_movcond_br1(s, cond, ret, cmp1, cmp2, val2, c_val2);
1430        return;
1431    }
1432
1433    tmp = (ret == cmp1 || ret == cmp2 ? TCG_REG_TMP1 : ret);
1434    if (c_val1) {
1435        tcg_out_movi(s, TCG_TYPE_REG, tmp, val1);
1436    } else {
1437        tcg_out_mov(s, TCG_TYPE_REG, tmp, val1);
1438    }
1439    tcg_out_movcond_br1(s, cond, tmp, cmp1, cmp2, val2, c_val2);
1440    tcg_out_mov(s, TCG_TYPE_REG, ret, tmp);
1441}
1442
1443static void tcg_out_movcond(TCGContext *s, TCGType type, TCGCond cond,
1444                            TCGReg ret, TCGReg cmp1, TCGArg cmp2, bool c_cmp2,
1445                            TCGArg val1, bool c_val1,
1446                            TCGArg val2, bool c_val2)
1447{
1448    int tmpflags;
1449    TCGReg t;
1450
1451    if (!(cpuinfo & CPUINFO_ZICOND) && (!c_cmp2 || cmp2 == 0)) {
1452        tcg_out_movcond_br2(s, cond, ret, cmp1, cmp2,
1453                            val1, c_val1, val2, c_val2);
1454        return;
1455    }
1456
1457    tmpflags = tcg_out_setcond_int(s, cond, TCG_REG_TMP0, cmp1, cmp2, c_cmp2);
1458    t = tmpflags & ~SETCOND_FLAGS;
1459
1460    if (cpuinfo & CPUINFO_ZICOND) {
1461        if (tmpflags & SETCOND_INV) {
1462            tcg_out_movcond_zicond(s, ret, t, val2, c_val2, val1, c_val1);
1463        } else {
1464            tcg_out_movcond_zicond(s, ret, t, val1, c_val1, val2, c_val2);
1465        }
1466    } else {
1467        cond = tmpflags & SETCOND_INV ? TCG_COND_EQ : TCG_COND_NE;
1468        tcg_out_movcond_br2(s, cond, ret, t, TCG_REG_ZERO,
1469                            val1, c_val1, val2, c_val2);
1470    }
1471}
1472
1473static const TCGOutOpMovcond outop_movcond = {
1474    .base.static_constraint = C_O1_I4(r, r, rI, rM, rM),
1475    .out = tcg_out_movcond,
1476};
1477
1478static void tcg_out_cltz(TCGContext *s, TCGType type, RISCVInsn insn,
1479                         TCGReg ret, TCGReg src1, int src2, bool c_src2)
1480{
1481    tcg_out_opc_imm(s, insn, ret, src1, 0);
1482
1483    if (!c_src2 || src2 != (type == TCG_TYPE_I32 ? 32 : 64)) {
1484        /*
1485         * The requested zero result does not match the insn, so adjust.
1486         * Note that constraints put 'ret' in a new register, so the
1487         * computation above did not clobber either 'src1' or 'src2'.
1488         */
1489        tcg_out_movcond(s, type, TCG_COND_EQ, ret, src1, 0, true,
1490                        src2, c_src2, ret, false);
1491    }
1492}
1493
1494static void tcg_out_cmpsel(TCGContext *s, TCGType type, unsigned vece,
1495                           TCGCond cond, TCGReg ret,
1496                           TCGReg cmp1, TCGReg cmp2, bool c_cmp2,
1497                           TCGReg val1, bool c_val1,
1498                           TCGReg val2, bool c_val2)
1499{
1500    set_vtype_len_sew(s, type, vece);
1501
1502    /* Use only vmerge_vim if possible, by inverting the test. */
1503    if (c_val2 && !c_val1) {
1504        TCGArg temp = val1;
1505        cond = tcg_invert_cond(cond);
1506        val1 = val2;
1507        val2 = temp;
1508        c_val1 = true;
1509        c_val2 = false;
1510    }
1511
1512    /* Perform the comparison into V0 mask. */
1513    if (c_cmp2) {
1514        tcg_out_opc_vi(s, tcg_cmpcond_to_rvv_vi[cond].op, TCG_REG_V0, cmp1,
1515                       cmp2 - tcg_cmpcond_to_rvv_vi[cond].adjust);
1516    } else if (tcg_cmpcond_to_rvv_vv[cond].swap) {
1517        tcg_out_opc_vv(s, tcg_cmpcond_to_rvv_vv[cond].op,
1518                       TCG_REG_V0, cmp2, cmp1);
1519    } else {
1520        tcg_out_opc_vv(s, tcg_cmpcond_to_rvv_vv[cond].op,
1521                       TCG_REG_V0, cmp1, cmp2);
1522    }
1523    if (c_val1) {
1524        if (c_val2) {
1525            tcg_out_opc_vi(s, OPC_VMV_V_I, ret, 0, val2);
1526            val2 = ret;
1527        }
1528        /* vd[i] == v0.mask[i] ? imm : vs2[i] */
1529        tcg_out_opc_vim_mask(s, OPC_VMERGE_VIM, ret, val2, val1);
1530    } else {
1531        /* vd[i] == v0.mask[i] ? vs1[i] : vs2[i] */
1532        tcg_out_opc_vvm_mask(s, OPC_VMERGE_VVM, ret, val2, val1);
1533    }
1534}
1535
1536static void tcg_out_vshifti(TCGContext *s, RISCVInsn opc_vi, RISCVInsn opc_vx,
1537                             TCGReg dst, TCGReg src, unsigned imm)
1538{
1539    if (imm < 32) {
1540        tcg_out_opc_vi(s, opc_vi, dst, src, imm);
1541    } else {
1542        tcg_out_movi(s, TCG_TYPE_I32, TCG_REG_TMP0, imm);
1543        tcg_out_opc_vx(s, opc_vx, dst, src, TCG_REG_TMP0);
1544    }
1545}
1546
1547static void init_setting_vtype(TCGContext *s)
1548{
1549    s->riscv_cur_type = TCG_TYPE_COUNT;
1550}
1551
1552static void tcg_out_call_int(TCGContext *s, const tcg_insn_unit *arg, bool tail)
1553{
1554    TCGReg link = tail ? TCG_REG_ZERO : TCG_REG_RA;
1555    ptrdiff_t offset = tcg_pcrel_diff(s, arg);
1556    int ret;
1557
1558    init_setting_vtype(s);
1559
1560    tcg_debug_assert((offset & 1) == 0);
1561    if (offset == sextreg(offset, 0, 20)) {
1562        /* short jump: -2097150 to 2097152 */
1563        tcg_out_opc_jump(s, OPC_JAL, link, offset);
1564    } else if (offset == (int32_t)offset) {
1565        /* long jump: -2147483646 to 2147483648 */
1566        tcg_out_opc_upper(s, OPC_AUIPC, TCG_REG_TMP0, 0);
1567        tcg_out_opc_imm(s, OPC_JALR, link, TCG_REG_TMP0, 0);
1568        ret = reloc_call(s->code_ptr - 2, arg);
1569        tcg_debug_assert(ret == true);
1570    } else {
1571        /* far jump: 64-bit */
1572        tcg_target_long imm = sextreg((tcg_target_long)arg, 0, 12);
1573        tcg_target_long base = (tcg_target_long)arg - imm;
1574        tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_TMP0, base);
1575        tcg_out_opc_imm(s, OPC_JALR, link, TCG_REG_TMP0, imm);
1576    }
1577}
1578
1579static void tcg_out_call(TCGContext *s, const tcg_insn_unit *arg,
1580                         const TCGHelperInfo *info)
1581{
1582    tcg_out_call_int(s, arg, false);
1583}
1584
1585static void tcg_out_mb(TCGContext *s, TCGArg a0)
1586{
1587    tcg_insn_unit insn = OPC_FENCE;
1588
1589    if (a0 & TCG_MO_LD_LD) {
1590        insn |= 0x02200000;
1591    }
1592    if (a0 & TCG_MO_ST_LD) {
1593        insn |= 0x01200000;
1594    }
1595    if (a0 & TCG_MO_LD_ST) {
1596        insn |= 0x02100000;
1597    }
1598    if (a0 & TCG_MO_ST_ST) {
1599        insn |= 0x01100000;
1600    }
1601    tcg_out32(s, insn);
1602}
1603
1604/*
1605 * Load/store and TLB
1606 */
1607
1608static void tcg_out_goto(TCGContext *s, const tcg_insn_unit *target)
1609{
1610    tcg_out_opc_jump(s, OPC_JAL, TCG_REG_ZERO, 0);
1611    bool ok = reloc_jimm20(s->code_ptr - 1, target);
1612    tcg_debug_assert(ok);
1613}
1614
1615bool tcg_target_has_memory_bswap(MemOp memop)
1616{
1617    return false;
1618}
1619
1620/* We have three temps, we might as well expose them. */
1621static const TCGLdstHelperParam ldst_helper_param = {
1622    .ntmp = 3, .tmp = { TCG_REG_TMP0, TCG_REG_TMP1, TCG_REG_TMP2 }
1623};
1624
1625static bool tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *l)
1626{
1627    MemOp opc = get_memop(l->oi);
1628
1629    /* resolve label address */
1630    if (!reloc_sbimm12(l->label_ptr[0], tcg_splitwx_to_rx(s->code_ptr))) {
1631        return false;
1632    }
1633
1634    /* call load helper */
1635    tcg_out_ld_helper_args(s, l, &ldst_helper_param);
1636    tcg_out_call_int(s, qemu_ld_helpers[opc & MO_SSIZE], false);
1637    tcg_out_ld_helper_ret(s, l, true, &ldst_helper_param);
1638
1639    tcg_out_goto(s, l->raddr);
1640    return true;
1641}
1642
1643static bool tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *l)
1644{
1645    MemOp opc = get_memop(l->oi);
1646
1647    /* resolve label address */
1648    if (!reloc_sbimm12(l->label_ptr[0], tcg_splitwx_to_rx(s->code_ptr))) {
1649        return false;
1650    }
1651
1652    /* call store helper */
1653    tcg_out_st_helper_args(s, l, &ldst_helper_param);
1654    tcg_out_call_int(s, qemu_st_helpers[opc & MO_SIZE], false);
1655
1656    tcg_out_goto(s, l->raddr);
1657    return true;
1658}
1659
1660/* We expect to use a 12-bit negative offset from ENV.  */
1661#define MIN_TLB_MASK_TABLE_OFS  -(1 << 11)
1662
1663/*
1664 * For system-mode, perform the TLB load and compare.
1665 * For user-mode, perform any required alignment tests.
1666 * In both cases, return a TCGLabelQemuLdst structure if the slow path
1667 * is required and fill in @h with the host address for the fast path.
1668 */
1669static TCGLabelQemuLdst *prepare_host_addr(TCGContext *s, TCGReg *pbase,
1670                                           TCGReg addr_reg, MemOpIdx oi,
1671                                           bool is_ld)
1672{
1673    TCGType addr_type = s->addr_type;
1674    TCGLabelQemuLdst *ldst = NULL;
1675    MemOp opc = get_memop(oi);
1676    TCGAtomAlign aa;
1677    unsigned a_mask;
1678
1679    aa = atom_and_align_for_opc(s, opc, MO_ATOM_IFALIGN, false);
1680    a_mask = (1u << aa.align) - 1;
1681
1682    if (tcg_use_softmmu) {
1683        unsigned s_bits = opc & MO_SIZE;
1684        unsigned s_mask = (1u << s_bits) - 1;
1685        int mem_index = get_mmuidx(oi);
1686        int fast_ofs = tlb_mask_table_ofs(s, mem_index);
1687        int mask_ofs = fast_ofs + offsetof(CPUTLBDescFast, mask);
1688        int table_ofs = fast_ofs + offsetof(CPUTLBDescFast, table);
1689        int compare_mask;
1690        TCGReg addr_adj;
1691
1692        ldst = new_ldst_label(s);
1693        ldst->is_ld = is_ld;
1694        ldst->oi = oi;
1695        ldst->addr_reg = addr_reg;
1696
1697        init_setting_vtype(s);
1698
1699        tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_TMP0, TCG_AREG0, mask_ofs);
1700        tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_TMP1, TCG_AREG0, table_ofs);
1701
1702        tcg_out_opc_imm(s, OPC_SRLI, TCG_REG_TMP2, addr_reg,
1703                        s->page_bits - CPU_TLB_ENTRY_BITS);
1704        tcg_out_opc_reg(s, OPC_AND, TCG_REG_TMP2, TCG_REG_TMP2, TCG_REG_TMP0);
1705        tcg_out_opc_reg(s, OPC_ADD, TCG_REG_TMP2, TCG_REG_TMP2, TCG_REG_TMP1);
1706
1707        /*
1708         * For aligned accesses, we check the first byte and include the
1709         * alignment bits within the address.  For unaligned access, we
1710         * check that we don't cross pages using the address of the last
1711         * byte of the access.
1712         */
1713        addr_adj = addr_reg;
1714        if (a_mask < s_mask) {
1715            addr_adj = TCG_REG_TMP0;
1716            tcg_out_opc_imm(s, addr_type == TCG_TYPE_I32 ? OPC_ADDIW : OPC_ADDI,
1717                            addr_adj, addr_reg, s_mask - a_mask);
1718        }
1719        compare_mask = s->page_mask | a_mask;
1720        if (compare_mask == sextreg(compare_mask, 0, 12)) {
1721            tcg_out_opc_imm(s, OPC_ANDI, TCG_REG_TMP1, addr_adj, compare_mask);
1722        } else {
1723            tcg_out_movi(s, addr_type, TCG_REG_TMP1, compare_mask);
1724            tcg_out_opc_reg(s, OPC_AND, TCG_REG_TMP1, TCG_REG_TMP1, addr_adj);
1725        }
1726
1727        /* Load the tlb comparator and the addend.  */
1728        QEMU_BUILD_BUG_ON(HOST_BIG_ENDIAN);
1729        tcg_out_ld(s, addr_type, TCG_REG_TMP0, TCG_REG_TMP2,
1730                   is_ld ? offsetof(CPUTLBEntry, addr_read)
1731                         : offsetof(CPUTLBEntry, addr_write));
1732        tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_TMP2, TCG_REG_TMP2,
1733                   offsetof(CPUTLBEntry, addend));
1734
1735        /* Compare masked address with the TLB entry. */
1736        ldst->label_ptr[0] = s->code_ptr;
1737        tcg_out_opc_branch(s, OPC_BNE, TCG_REG_TMP0, TCG_REG_TMP1, 0);
1738
1739        /* TLB Hit - translate address using addend.  */
1740        if (addr_type != TCG_TYPE_I32) {
1741            tcg_out_opc_reg(s, OPC_ADD, TCG_REG_TMP0, addr_reg, TCG_REG_TMP2);
1742        } else if (cpuinfo & CPUINFO_ZBA) {
1743            tcg_out_opc_reg(s, OPC_ADD_UW, TCG_REG_TMP0,
1744                            addr_reg, TCG_REG_TMP2);
1745        } else {
1746            tcg_out_ext32u(s, TCG_REG_TMP0, addr_reg);
1747            tcg_out_opc_reg(s, OPC_ADD, TCG_REG_TMP0,
1748                            TCG_REG_TMP0, TCG_REG_TMP2);
1749        }
1750        *pbase = TCG_REG_TMP0;
1751    } else {
1752        TCGReg base;
1753
1754        if (a_mask) {
1755            ldst = new_ldst_label(s);
1756            ldst->is_ld = is_ld;
1757            ldst->oi = oi;
1758            ldst->addr_reg = addr_reg;
1759
1760            init_setting_vtype(s);
1761
1762            /* We are expecting alignment max 7, so we can always use andi. */
1763            tcg_debug_assert(a_mask == sextreg(a_mask, 0, 12));
1764            tcg_out_opc_imm(s, OPC_ANDI, TCG_REG_TMP1, addr_reg, a_mask);
1765
1766            ldst->label_ptr[0] = s->code_ptr;
1767            tcg_out_opc_branch(s, OPC_BNE, TCG_REG_TMP1, TCG_REG_ZERO, 0);
1768        }
1769
1770        if (guest_base != 0) {
1771            base = TCG_REG_TMP0;
1772            if (addr_type != TCG_TYPE_I32) {
1773                tcg_out_opc_reg(s, OPC_ADD, base, addr_reg,
1774                                TCG_GUEST_BASE_REG);
1775            } else if (cpuinfo & CPUINFO_ZBA) {
1776                tcg_out_opc_reg(s, OPC_ADD_UW, base, addr_reg,
1777                                TCG_GUEST_BASE_REG);
1778            } else {
1779                tcg_out_ext32u(s, base, addr_reg);
1780                tcg_out_opc_reg(s, OPC_ADD, base, base, TCG_GUEST_BASE_REG);
1781            }
1782        } else if (addr_type != TCG_TYPE_I32) {
1783            base = addr_reg;
1784        } else {
1785            base = TCG_REG_TMP0;
1786            tcg_out_ext32u(s, base, addr_reg);
1787        }
1788        *pbase = base;
1789    }
1790
1791    return ldst;
1792}
1793
1794static void tcg_out_qemu_ld_direct(TCGContext *s, TCGReg val,
1795                                   TCGReg base, MemOp opc, TCGType type)
1796{
1797    /* Byte swapping is left to middle-end expansion. */
1798    tcg_debug_assert((opc & MO_BSWAP) == 0);
1799
1800    switch (opc & (MO_SSIZE)) {
1801    case MO_UB:
1802        tcg_out_opc_imm(s, OPC_LBU, val, base, 0);
1803        break;
1804    case MO_SB:
1805        tcg_out_opc_imm(s, OPC_LB, val, base, 0);
1806        break;
1807    case MO_UW:
1808        tcg_out_opc_imm(s, OPC_LHU, val, base, 0);
1809        break;
1810    case MO_SW:
1811        tcg_out_opc_imm(s, OPC_LH, val, base, 0);
1812        break;
1813    case MO_UL:
1814        if (type == TCG_TYPE_I64) {
1815            tcg_out_opc_imm(s, OPC_LWU, val, base, 0);
1816            break;
1817        }
1818        /* FALLTHRU */
1819    case MO_SL:
1820        tcg_out_opc_imm(s, OPC_LW, val, base, 0);
1821        break;
1822    case MO_UQ:
1823        tcg_out_opc_imm(s, OPC_LD, val, base, 0);
1824        break;
1825    default:
1826        g_assert_not_reached();
1827    }
1828}
1829
1830static void tcg_out_qemu_ld(TCGContext *s, TCGReg data_reg, TCGReg addr_reg,
1831                            MemOpIdx oi, TCGType data_type)
1832{
1833    TCGLabelQemuLdst *ldst;
1834    TCGReg base;
1835
1836    ldst = prepare_host_addr(s, &base, addr_reg, oi, true);
1837    tcg_out_qemu_ld_direct(s, data_reg, base, get_memop(oi), data_type);
1838
1839    if (ldst) {
1840        ldst->type = data_type;
1841        ldst->datalo_reg = data_reg;
1842        ldst->raddr = tcg_splitwx_to_rx(s->code_ptr);
1843    }
1844}
1845
1846static void tcg_out_qemu_st_direct(TCGContext *s, TCGReg val,
1847                                   TCGReg base, MemOp opc)
1848{
1849    /* Byte swapping is left to middle-end expansion. */
1850    tcg_debug_assert((opc & MO_BSWAP) == 0);
1851
1852    switch (opc & (MO_SSIZE)) {
1853    case MO_8:
1854        tcg_out_opc_store(s, OPC_SB, base, val, 0);
1855        break;
1856    case MO_16:
1857        tcg_out_opc_store(s, OPC_SH, base, val, 0);
1858        break;
1859    case MO_32:
1860        tcg_out_opc_store(s, OPC_SW, base, val, 0);
1861        break;
1862    case MO_64:
1863        tcg_out_opc_store(s, OPC_SD, base, val, 0);
1864        break;
1865    default:
1866        g_assert_not_reached();
1867    }
1868}
1869
1870static void tcg_out_qemu_st(TCGContext *s, TCGReg data_reg, TCGReg addr_reg,
1871                            MemOpIdx oi, TCGType data_type)
1872{
1873    TCGLabelQemuLdst *ldst;
1874    TCGReg base;
1875
1876    ldst = prepare_host_addr(s, &base, addr_reg, oi, false);
1877    tcg_out_qemu_st_direct(s, data_reg, base, get_memop(oi));
1878
1879    if (ldst) {
1880        ldst->type = data_type;
1881        ldst->datalo_reg = data_reg;
1882        ldst->raddr = tcg_splitwx_to_rx(s->code_ptr);
1883    }
1884}
1885
1886static const tcg_insn_unit *tb_ret_addr;
1887
1888static void tcg_out_exit_tb(TCGContext *s, uintptr_t a0)
1889{
1890    /* Reuse the zeroing that exists for goto_ptr.  */
1891    if (a0 == 0) {
1892        tcg_out_call_int(s, tcg_code_gen_epilogue, true);
1893    } else {
1894        tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_A0, a0);
1895        tcg_out_call_int(s, tb_ret_addr, true);
1896    }
1897}
1898
1899static void tcg_out_goto_tb(TCGContext *s, int which)
1900{
1901    /* Direct branch will be patched by tb_target_set_jmp_target. */
1902    set_jmp_insn_offset(s, which);
1903    tcg_out32(s, OPC_JAL);
1904
1905    /* When branch is out of range, fall through to indirect. */
1906    tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_TMP0, TCG_REG_ZERO,
1907               get_jmp_target_addr(s, which));
1908    tcg_out_opc_imm(s, OPC_JALR, TCG_REG_ZERO, TCG_REG_TMP0, 0);
1909    set_jmp_reset_offset(s, which);
1910}
1911
1912void tb_target_set_jmp_target(const TranslationBlock *tb, int n,
1913                              uintptr_t jmp_rx, uintptr_t jmp_rw)
1914{
1915    uintptr_t addr = tb->jmp_target_addr[n];
1916    ptrdiff_t offset = addr - jmp_rx;
1917    tcg_insn_unit insn;
1918
1919    /* Either directly branch, or fall through to indirect branch. */
1920    if (offset == sextreg(offset, 0, 20)) {
1921        insn = encode_uj(OPC_JAL, TCG_REG_ZERO, offset);
1922    } else {
1923        insn = OPC_NOP;
1924    }
1925    qatomic_set((uint32_t *)jmp_rw, insn);
1926    flush_idcache_range(jmp_rx, jmp_rw, 4);
1927}
1928
1929
1930static void tgen_add(TCGContext *s, TCGType type,
1931                     TCGReg a0, TCGReg a1, TCGReg a2)
1932{
1933    RISCVInsn insn = type == TCG_TYPE_I32 ? OPC_ADDW : OPC_ADD;
1934    tcg_out_opc_reg(s, insn, a0, a1, a2);
1935}
1936
1937static void tgen_addi(TCGContext *s, TCGType type,
1938                      TCGReg a0, TCGReg a1, tcg_target_long a2)
1939{
1940    RISCVInsn insn = type == TCG_TYPE_I32 ? OPC_ADDIW : OPC_ADDI;
1941    tcg_out_opc_imm(s, insn, a0, a1, a2);
1942}
1943
1944static const TCGOutOpBinary outop_add = {
1945    .base.static_constraint = C_O1_I2(r, r, rI),
1946    .out_rrr = tgen_add,
1947    .out_rri = tgen_addi,
1948};
1949
1950static void tgen_and(TCGContext *s, TCGType type,
1951                     TCGReg a0, TCGReg a1, TCGReg a2)
1952{
1953    tcg_out_opc_reg(s, OPC_AND, a0, a1, a2);
1954}
1955
1956static void tgen_andi(TCGContext *s, TCGType type,
1957                      TCGReg a0, TCGReg a1, tcg_target_long a2)
1958{
1959    tcg_out_opc_imm(s, OPC_ANDI, a0, a1, a2);
1960}
1961
1962static const TCGOutOpBinary outop_and = {
1963    .base.static_constraint = C_O1_I2(r, r, rI),
1964    .out_rrr = tgen_and,
1965    .out_rri = tgen_andi,
1966};
1967
1968static void tgen_andc(TCGContext *s, TCGType type,
1969                      TCGReg a0, TCGReg a1, TCGReg a2)
1970{
1971    tcg_out_opc_reg(s, OPC_ANDN, a0, a1, a2);
1972}
1973
1974static TCGConstraintSetIndex cset_zbb_rrr(TCGType type, unsigned flags)
1975{
1976    return cpuinfo & CPUINFO_ZBB ? C_O1_I2(r, r, r) : C_NotImplemented;
1977}
1978
1979static const TCGOutOpBinary outop_andc = {
1980    .base.static_constraint = C_Dynamic,
1981    .base.dynamic_constraint = cset_zbb_rrr,
1982    .out_rrr = tgen_andc,
1983};
1984
1985static void tgen_clz(TCGContext *s, TCGType type,
1986                     TCGReg a0, TCGReg a1, TCGReg a2)
1987{
1988    RISCVInsn insn = type == TCG_TYPE_I32 ? OPC_CLZW : OPC_CLZ;
1989    tcg_out_cltz(s, type, insn, a0, a1, a2, false);
1990}
1991
1992static void tgen_clzi(TCGContext *s, TCGType type,
1993                      TCGReg a0, TCGReg a1, tcg_target_long a2)
1994{
1995    RISCVInsn insn = type == TCG_TYPE_I32 ? OPC_CLZW : OPC_CLZ;
1996    tcg_out_cltz(s, type, insn, a0, a1, a2, true);
1997}
1998
1999static TCGConstraintSetIndex cset_clzctz(TCGType type, unsigned flags)
2000{
2001    return cpuinfo & CPUINFO_ZBB ? C_N1_I2(r, r, rM) : C_NotImplemented;
2002}
2003
2004static const TCGOutOpBinary outop_clz = {
2005    .base.static_constraint = C_Dynamic,
2006    .base.dynamic_constraint = cset_clzctz,
2007    .out_rrr = tgen_clz,
2008    .out_rri = tgen_clzi,
2009};
2010
2011static void tgen_ctpop(TCGContext *s, TCGType type, TCGReg a0, TCGReg a1)
2012{
2013    RISCVInsn insn = type == TCG_TYPE_I32 ? OPC_CPOPW : OPC_CPOP;
2014    tcg_out_opc_imm(s, insn, a0, a1, 0);
2015}
2016
2017static TCGConstraintSetIndex cset_ctpop(TCGType type, unsigned flags)
2018{
2019    return cpuinfo & CPUINFO_ZBB ? C_O1_I1(r, r) : C_NotImplemented;
2020}
2021
2022static const TCGOutOpUnary outop_ctpop = {
2023    .base.static_constraint = C_Dynamic,
2024    .base.dynamic_constraint = cset_ctpop,
2025    .out_rr = tgen_ctpop,
2026};
2027
2028static void tgen_ctz(TCGContext *s, TCGType type,
2029                     TCGReg a0, TCGReg a1, TCGReg a2)
2030{
2031    RISCVInsn insn = type == TCG_TYPE_I32 ? OPC_CTZW : OPC_CTZ;
2032    tcg_out_cltz(s, type, insn, a0, a1, a2, false);
2033}
2034
2035static void tgen_ctzi(TCGContext *s, TCGType type,
2036                      TCGReg a0, TCGReg a1, tcg_target_long a2)
2037{
2038    RISCVInsn insn = type == TCG_TYPE_I32 ? OPC_CTZW : OPC_CTZ;
2039    tcg_out_cltz(s, type, insn, a0, a1, a2, true);
2040}
2041
2042static const TCGOutOpBinary outop_ctz = {
2043    .base.static_constraint = C_Dynamic,
2044    .base.dynamic_constraint = cset_clzctz,
2045    .out_rrr = tgen_ctz,
2046    .out_rri = tgen_ctzi,
2047};
2048
2049static void tgen_divs(TCGContext *s, TCGType type,
2050                      TCGReg a0, TCGReg a1, TCGReg a2)
2051{
2052    RISCVInsn insn = type == TCG_TYPE_I32 ? OPC_DIVW : OPC_DIV;
2053    tcg_out_opc_reg(s, insn, a0, a1, a2);
2054}
2055
2056static const TCGOutOpBinary outop_divs = {
2057    .base.static_constraint = C_O1_I2(r, r, r),
2058    .out_rrr = tgen_divs,
2059};
2060
2061static const TCGOutOpDivRem outop_divs2 = {
2062    .base.static_constraint = C_NotImplemented,
2063};
2064
2065static void tgen_divu(TCGContext *s, TCGType type,
2066                      TCGReg a0, TCGReg a1, TCGReg a2)
2067{
2068    RISCVInsn insn = type == TCG_TYPE_I32 ? OPC_DIVUW : OPC_DIVU;
2069    tcg_out_opc_reg(s, insn, a0, a1, a2);
2070}
2071
2072static const TCGOutOpBinary outop_divu = {
2073    .base.static_constraint = C_O1_I2(r, r, r),
2074    .out_rrr = tgen_divu,
2075};
2076
2077static const TCGOutOpDivRem outop_divu2 = {
2078    .base.static_constraint = C_NotImplemented,
2079};
2080
2081static void tgen_eqv(TCGContext *s, TCGType type,
2082                     TCGReg a0, TCGReg a1, TCGReg a2)
2083{
2084    tcg_out_opc_reg(s, OPC_XNOR, a0, a1, a2);
2085}
2086
2087static const TCGOutOpBinary outop_eqv = {
2088    .base.static_constraint = C_Dynamic,
2089    .base.dynamic_constraint = cset_zbb_rrr,
2090    .out_rrr = tgen_eqv,
2091};
2092
2093static void tgen_extrh_i64_i32(TCGContext *s, TCGType t, TCGReg a0, TCGReg a1)
2094{
2095    tcg_out_opc_imm(s, OPC_SRAI, a0, a1, 32);
2096}
2097
2098static const TCGOutOpUnary outop_extrh_i64_i32 = {
2099    .base.static_constraint = C_O1_I1(r, r),
2100    .out_rr = tgen_extrh_i64_i32,
2101};
2102
2103static void tgen_mul(TCGContext *s, TCGType type,
2104                     TCGReg a0, TCGReg a1, TCGReg a2)
2105{
2106    RISCVInsn insn = type == TCG_TYPE_I32 ? OPC_MULW : OPC_MUL;
2107    tcg_out_opc_reg(s, insn, a0, a1, a2);
2108}
2109
2110static const TCGOutOpBinary outop_mul = {
2111    .base.static_constraint = C_O1_I2(r, r, r),
2112    .out_rrr = tgen_mul,
2113};
2114
2115static const TCGOutOpMul2 outop_muls2 = {
2116    .base.static_constraint = C_NotImplemented,
2117};
2118
2119static TCGConstraintSetIndex cset_mulh(TCGType type, unsigned flags)
2120{
2121    return type == TCG_TYPE_I32 ? C_NotImplemented : C_O1_I2(r, r, r);
2122}
2123
2124static void tgen_mulsh(TCGContext *s, TCGType type,
2125                       TCGReg a0, TCGReg a1, TCGReg a2)
2126{
2127    tcg_out_opc_reg(s, OPC_MULH, a0, a1, a2);
2128}
2129
2130static const TCGOutOpBinary outop_mulsh = {
2131    .base.static_constraint = C_Dynamic,
2132    .base.dynamic_constraint = cset_mulh,
2133    .out_rrr = tgen_mulsh,
2134};
2135
2136static const TCGOutOpMul2 outop_mulu2 = {
2137    .base.static_constraint = C_NotImplemented,
2138};
2139
2140static void tgen_muluh(TCGContext *s, TCGType type,
2141                       TCGReg a0, TCGReg a1, TCGReg a2)
2142{
2143    tcg_out_opc_reg(s, OPC_MULHU, a0, a1, a2);
2144}
2145
2146static const TCGOutOpBinary outop_muluh = {
2147    .base.static_constraint = C_Dynamic,
2148    .base.dynamic_constraint = cset_mulh,
2149    .out_rrr = tgen_muluh,
2150};
2151
2152static const TCGOutOpBinary outop_nand = {
2153    .base.static_constraint = C_NotImplemented,
2154};
2155
2156static const TCGOutOpBinary outop_nor = {
2157    .base.static_constraint = C_NotImplemented,
2158};
2159
2160static void tgen_or(TCGContext *s, TCGType type,
2161                    TCGReg a0, TCGReg a1, TCGReg a2)
2162{
2163    tcg_out_opc_reg(s, OPC_OR, a0, a1, a2);
2164}
2165
2166static void tgen_ori(TCGContext *s, TCGType type,
2167                     TCGReg a0, TCGReg a1, tcg_target_long a2)
2168{
2169    tcg_out_opc_imm(s, OPC_ORI, a0, a1, a2);
2170}
2171
2172static const TCGOutOpBinary outop_or = {
2173    .base.static_constraint = C_O1_I2(r, r, rI),
2174    .out_rrr = tgen_or,
2175    .out_rri = tgen_ori,
2176};
2177
2178static void tgen_orc(TCGContext *s, TCGType type,
2179                     TCGReg a0, TCGReg a1, TCGReg a2)
2180{
2181    tcg_out_opc_reg(s, OPC_ORN, a0, a1, a2);
2182}
2183
2184static const TCGOutOpBinary outop_orc = {
2185    .base.static_constraint = C_Dynamic,
2186    .base.dynamic_constraint = cset_zbb_rrr,
2187    .out_rrr = tgen_orc,
2188};
2189
2190static void tgen_rems(TCGContext *s, TCGType type,
2191                      TCGReg a0, TCGReg a1, TCGReg a2)
2192{
2193    RISCVInsn insn = type == TCG_TYPE_I32 ? OPC_REMW : OPC_REM;
2194    tcg_out_opc_reg(s, insn, a0, a1, a2);
2195}
2196
2197static const TCGOutOpBinary outop_rems = {
2198    .base.static_constraint = C_O1_I2(r, r, r),
2199    .out_rrr = tgen_rems,
2200};
2201
2202static void tgen_remu(TCGContext *s, TCGType type,
2203                      TCGReg a0, TCGReg a1, TCGReg a2)
2204{
2205    RISCVInsn insn = type == TCG_TYPE_I32 ? OPC_REMUW : OPC_REMU;
2206    tcg_out_opc_reg(s, insn, a0, a1, a2);
2207}
2208
2209static const TCGOutOpBinary outop_remu = {
2210    .base.static_constraint = C_O1_I2(r, r, r),
2211    .out_rrr = tgen_remu,
2212};
2213
2214static TCGConstraintSetIndex cset_rot(TCGType type, unsigned flags)
2215{
2216    return cpuinfo & CPUINFO_ZBB ? C_O1_I2(r, r, ri) : C_NotImplemented;
2217}
2218
2219static void tgen_rotr(TCGContext *s, TCGType type,
2220                      TCGReg a0, TCGReg a1, TCGReg a2)
2221{
2222    RISCVInsn insn = type == TCG_TYPE_I32 ? OPC_RORW : OPC_ROR;
2223    tcg_out_opc_reg(s, insn, a0, a1, a2);
2224}
2225
2226static void tgen_rotri(TCGContext *s, TCGType type,
2227                       TCGReg a0, TCGReg a1, tcg_target_long a2)
2228{
2229    RISCVInsn insn = type == TCG_TYPE_I32 ? OPC_RORIW : OPC_RORI;
2230    unsigned mask = type == TCG_TYPE_I32 ? 31 : 63;
2231    tcg_out_opc_imm(s, insn, a0, a1, a2 & mask);
2232}
2233
2234static const TCGOutOpBinary outop_rotr = {
2235    .base.static_constraint = C_Dynamic,
2236    .base.dynamic_constraint = cset_rot,
2237    .out_rrr = tgen_rotr,
2238    .out_rri = tgen_rotri,
2239};
2240
2241static void tgen_rotl(TCGContext *s, TCGType type,
2242                      TCGReg a0, TCGReg a1, TCGReg a2)
2243{
2244    RISCVInsn insn = type == TCG_TYPE_I32 ? OPC_ROLW : OPC_ROL;
2245    tcg_out_opc_reg(s, insn, a0, a1, a2);
2246}
2247
2248static void tgen_rotli(TCGContext *s, TCGType type,
2249                       TCGReg a0, TCGReg a1, tcg_target_long a2)
2250{
2251    tgen_rotri(s, type, a0, a1, -a2);
2252}
2253
2254static const TCGOutOpBinary outop_rotl = {
2255    .base.static_constraint = C_Dynamic,
2256    .base.dynamic_constraint = cset_rot,
2257    .out_rrr = tgen_rotl,
2258    .out_rri = tgen_rotli,
2259};
2260
2261static void tgen_sar(TCGContext *s, TCGType type,
2262                     TCGReg a0, TCGReg a1, TCGReg a2)
2263{
2264    RISCVInsn insn = type == TCG_TYPE_I32 ? OPC_SRAW : OPC_SRA;
2265    tcg_out_opc_reg(s, insn, a0, a1, a2);
2266}
2267
2268static void tgen_sari(TCGContext *s, TCGType type,
2269                      TCGReg a0, TCGReg a1, tcg_target_long a2)
2270{
2271    RISCVInsn insn = type == TCG_TYPE_I32 ? OPC_SRAIW : OPC_SRAI;
2272    unsigned mask = type == TCG_TYPE_I32 ? 31 : 63;
2273    tcg_out_opc_imm(s, insn, a0, a1, a2 & mask);
2274}
2275
2276static const TCGOutOpBinary outop_sar = {
2277    .base.static_constraint = C_O1_I2(r, r, ri),
2278    .out_rrr = tgen_sar,
2279    .out_rri = tgen_sari,
2280};
2281
2282static void tgen_shl(TCGContext *s, TCGType type,
2283                     TCGReg a0, TCGReg a1, TCGReg a2)
2284{
2285    RISCVInsn insn = type == TCG_TYPE_I32 ? OPC_SLLW : OPC_SLL;
2286    tcg_out_opc_reg(s, insn, a0, a1, a2);
2287}
2288
2289static void tgen_shli(TCGContext *s, TCGType type,
2290                      TCGReg a0, TCGReg a1, tcg_target_long a2)
2291{
2292    RISCVInsn insn = type == TCG_TYPE_I32 ? OPC_SLLIW : OPC_SLLI;
2293    unsigned mask = type == TCG_TYPE_I32 ? 31 : 63;
2294    tcg_out_opc_imm(s, insn, a0, a1, a2 & mask);
2295}
2296
2297static const TCGOutOpBinary outop_shl = {
2298    .base.static_constraint = C_O1_I2(r, r, ri),
2299    .out_rrr = tgen_shl,
2300    .out_rri = tgen_shli,
2301};
2302
2303static void tgen_shr(TCGContext *s, TCGType type,
2304                     TCGReg a0, TCGReg a1, TCGReg a2)
2305{
2306    RISCVInsn insn = type == TCG_TYPE_I32 ? OPC_SRLW : OPC_SRL;
2307    tcg_out_opc_reg(s, insn, a0, a1, a2);
2308}
2309
2310static void tgen_shri(TCGContext *s, TCGType type,
2311                      TCGReg a0, TCGReg a1, tcg_target_long a2)
2312{
2313    RISCVInsn insn = type == TCG_TYPE_I32 ? OPC_SRLIW : OPC_SRLI;
2314    unsigned mask = type == TCG_TYPE_I32 ? 31 : 63;
2315    tcg_out_opc_imm(s, insn, a0, a1, a2 & mask);
2316}
2317
2318static const TCGOutOpBinary outop_shr = {
2319    .base.static_constraint = C_O1_I2(r, r, ri),
2320    .out_rrr = tgen_shr,
2321    .out_rri = tgen_shri,
2322};
2323
2324static void tgen_sub(TCGContext *s, TCGType type,
2325                     TCGReg a0, TCGReg a1, TCGReg a2)
2326{
2327    RISCVInsn insn = type == TCG_TYPE_I32 ? OPC_SUBW : OPC_SUB;
2328    tcg_out_opc_reg(s, insn, a0, a1, a2);
2329}
2330
2331static const TCGOutOpSubtract outop_sub = {
2332    .base.static_constraint = C_O1_I2(r, r, r),
2333    .out_rrr = tgen_sub,
2334};
2335
2336static void tgen_xor(TCGContext *s, TCGType type,
2337                     TCGReg a0, TCGReg a1, TCGReg a2)
2338{
2339    tcg_out_opc_reg(s, OPC_XOR, a0, a1, a2);
2340}
2341
2342static void tgen_xori(TCGContext *s, TCGType type,
2343                      TCGReg a0, TCGReg a1, tcg_target_long a2)
2344{
2345    tcg_out_opc_imm(s, OPC_XORI, a0, a1, a2);
2346}
2347
2348static const TCGOutOpBinary outop_xor = {
2349    .base.static_constraint = C_O1_I2(r, r, rI),
2350    .out_rrr = tgen_xor,
2351    .out_rri = tgen_xori,
2352};
2353
2354static TCGConstraintSetIndex cset_bswap(TCGType type, unsigned flags)
2355{
2356    return cpuinfo & CPUINFO_ZBB ? C_O1_I1(r, r) : C_NotImplemented;
2357}
2358
2359static void tgen_bswap16(TCGContext *s, TCGType type,
2360                         TCGReg a0, TCGReg a1, unsigned flags)
2361{
2362    tcg_out_opc_imm(s, OPC_REV8, a0, a1, 0);
2363    if (flags & TCG_BSWAP_OZ) {
2364        tcg_out_opc_imm(s, OPC_SRLI, a0, a0, 48);
2365    } else {
2366        tcg_out_opc_imm(s, OPC_SRAI, a0, a0, 48);
2367    }
2368}
2369
2370static const TCGOutOpBswap outop_bswap16 = {
2371    .base.static_constraint = C_Dynamic,
2372    .base.dynamic_constraint = cset_bswap,
2373    .out_rr = tgen_bswap16,
2374};
2375
2376static void tgen_bswap32(TCGContext *s, TCGType type,
2377                         TCGReg a0, TCGReg a1, unsigned flags)
2378{
2379    tcg_out_opc_imm(s, OPC_REV8, a0, a1, 0);
2380    if (flags & TCG_BSWAP_OZ) {
2381        tcg_out_opc_imm(s, OPC_SRLI, a0, a0, 32);
2382    } else {
2383        tcg_out_opc_imm(s, OPC_SRAI, a0, a0, 32);
2384    }
2385}
2386
2387static const TCGOutOpBswap outop_bswap32 = {
2388    .base.static_constraint = C_Dynamic,
2389    .base.dynamic_constraint = cset_bswap,
2390    .out_rr = tgen_bswap32,
2391};
2392
2393static void tgen_bswap64(TCGContext *s, TCGType type, TCGReg a0, TCGReg a1)
2394{
2395    tcg_out_opc_imm(s, OPC_REV8, a0, a1, 0);
2396}
2397
2398static const TCGOutOpUnary outop_bswap64 = {
2399    .base.static_constraint = C_Dynamic,
2400    .base.dynamic_constraint = cset_bswap,
2401    .out_rr = tgen_bswap64,
2402};
2403
2404static void tgen_neg(TCGContext *s, TCGType type, TCGReg a0, TCGReg a1)
2405{
2406    tgen_sub(s, type, a0, TCG_REG_ZERO, a1);
2407}
2408
2409static const TCGOutOpUnary outop_neg = {
2410    .base.static_constraint = C_O1_I1(r, r),
2411    .out_rr = tgen_neg,
2412};
2413
2414static void tgen_not(TCGContext *s, TCGType type, TCGReg a0, TCGReg a1)
2415{
2416    tgen_xori(s, type, a0, a1, -1);
2417}
2418
2419static const TCGOutOpUnary outop_not = {
2420    .base.static_constraint = C_O1_I1(r, r),
2421    .out_rr = tgen_not,
2422};
2423
2424static const TCGOutOpDeposit outop_deposit = {
2425    .base.static_constraint = C_NotImplemented,
2426};
2427
2428static void tgen_extract(TCGContext *s, TCGType type, TCGReg a0, TCGReg a1,
2429                         unsigned ofs, unsigned len)
2430{
2431    if (ofs == 0) {
2432        switch (len) {
2433        case 16:
2434            tcg_out_ext16u(s, a0, a1);
2435            return;
2436        case 32:
2437            tcg_out_ext32u(s, a0, a1);
2438            return;
2439        }
2440    }
2441    if (ofs + len == 32) {
2442        tgen_shli(s, TCG_TYPE_I32, a0, a1, ofs);
2443        return;
2444    }
2445    if (len == 1) {
2446        tcg_out_opc_imm(s, OPC_BEXTI, a0, a1, ofs);
2447        return;
2448    }
2449    g_assert_not_reached();
2450}
2451
2452static const TCGOutOpExtract outop_extract = {
2453    .base.static_constraint = C_O1_I1(r, r),
2454    .out_rr = tgen_extract,
2455};
2456
2457static void tgen_sextract(TCGContext *s, TCGType type, TCGReg a0, TCGReg a1,
2458                          unsigned ofs, unsigned len)
2459{
2460    if (ofs == 0) {
2461        switch (len) {
2462        case 8:
2463            tcg_out_ext8s(s, type, a0, a1);
2464            return;
2465        case 16:
2466            tcg_out_ext16s(s, type, a0, a1);
2467            return;
2468        case 32:
2469            tcg_out_ext32s(s, a0, a1);
2470            return;
2471        }
2472    } else if (ofs + len == 32) {
2473        tgen_sari(s, TCG_TYPE_I32, a0, a1, ofs);
2474        return;
2475    }
2476    g_assert_not_reached();
2477}
2478
2479static const TCGOutOpExtract outop_sextract = {
2480    .base.static_constraint = C_O1_I1(r, r),
2481    .out_rr = tgen_sextract,
2482};
2483
2484static const TCGOutOpExtract2 outop_extract2 = {
2485    .base.static_constraint = C_NotImplemented,
2486};
2487
2488
2489static void tcg_out_op(TCGContext *s, TCGOpcode opc, TCGType type,
2490                       const TCGArg args[TCG_MAX_OP_ARGS],
2491                       const int const_args[TCG_MAX_OP_ARGS])
2492{
2493    TCGArg a0 = args[0];
2494    TCGArg a1 = args[1];
2495    TCGArg a2 = args[2];
2496
2497    switch (opc) {
2498    case INDEX_op_goto_ptr:
2499        tcg_out_opc_imm(s, OPC_JALR, TCG_REG_ZERO, a0, 0);
2500        break;
2501
2502    case INDEX_op_br:
2503        tcg_out_reloc(s, s->code_ptr, R_RISCV_JAL, arg_label(a0), 0);
2504        tcg_out_opc_jump(s, OPC_JAL, TCG_REG_ZERO, 0);
2505        break;
2506
2507    case INDEX_op_ld8u_i32:
2508    case INDEX_op_ld8u_i64:
2509        tcg_out_ldst(s, OPC_LBU, a0, a1, a2);
2510        break;
2511    case INDEX_op_ld8s_i32:
2512    case INDEX_op_ld8s_i64:
2513        tcg_out_ldst(s, OPC_LB, a0, a1, a2);
2514        break;
2515    case INDEX_op_ld16u_i32:
2516    case INDEX_op_ld16u_i64:
2517        tcg_out_ldst(s, OPC_LHU, a0, a1, a2);
2518        break;
2519    case INDEX_op_ld16s_i32:
2520    case INDEX_op_ld16s_i64:
2521        tcg_out_ldst(s, OPC_LH, a0, a1, a2);
2522        break;
2523    case INDEX_op_ld32u_i64:
2524        tcg_out_ldst(s, OPC_LWU, a0, a1, a2);
2525        break;
2526    case INDEX_op_ld_i32:
2527    case INDEX_op_ld32s_i64:
2528        tcg_out_ldst(s, OPC_LW, a0, a1, a2);
2529        break;
2530    case INDEX_op_ld_i64:
2531        tcg_out_ldst(s, OPC_LD, a0, a1, a2);
2532        break;
2533
2534    case INDEX_op_st8_i32:
2535    case INDEX_op_st8_i64:
2536        tcg_out_ldst(s, OPC_SB, a0, a1, a2);
2537        break;
2538    case INDEX_op_st16_i32:
2539    case INDEX_op_st16_i64:
2540        tcg_out_ldst(s, OPC_SH, a0, a1, a2);
2541        break;
2542    case INDEX_op_st_i32:
2543    case INDEX_op_st32_i64:
2544        tcg_out_ldst(s, OPC_SW, a0, a1, a2);
2545        break;
2546    case INDEX_op_st_i64:
2547        tcg_out_ldst(s, OPC_SD, a0, a1, a2);
2548        break;
2549
2550    case INDEX_op_qemu_ld_i32:
2551        tcg_out_qemu_ld(s, a0, a1, a2, TCG_TYPE_I32);
2552        break;
2553    case INDEX_op_qemu_ld_i64:
2554        tcg_out_qemu_ld(s, a0, a1, a2, TCG_TYPE_I64);
2555        break;
2556    case INDEX_op_qemu_st_i32:
2557        tcg_out_qemu_st(s, a0, a1, a2, TCG_TYPE_I32);
2558        break;
2559    case INDEX_op_qemu_st_i64:
2560        tcg_out_qemu_st(s, a0, a1, a2, TCG_TYPE_I64);
2561        break;
2562
2563    case INDEX_op_mb:
2564        tcg_out_mb(s, a0);
2565        break;
2566
2567    case INDEX_op_call:     /* Always emitted via tcg_out_call.  */
2568    case INDEX_op_exit_tb:  /* Always emitted via tcg_out_exit_tb.  */
2569    case INDEX_op_goto_tb:  /* Always emitted via tcg_out_goto_tb.  */
2570    default:
2571        g_assert_not_reached();
2572    }
2573}
2574
2575static void tcg_out_vec_op(TCGContext *s, TCGOpcode opc,
2576                           unsigned vecl, unsigned vece,
2577                           const TCGArg args[TCG_MAX_OP_ARGS],
2578                           const int const_args[TCG_MAX_OP_ARGS])
2579{
2580    TCGType type = vecl + TCG_TYPE_V64;
2581    TCGArg a0, a1, a2;
2582    int c2;
2583
2584    a0 = args[0];
2585    a1 = args[1];
2586    a2 = args[2];
2587    c2 = const_args[2];
2588
2589    switch (opc) {
2590    case INDEX_op_dupm_vec:
2591        tcg_out_dupm_vec(s, type, vece, a0, a1, a2);
2592        break;
2593    case INDEX_op_ld_vec:
2594        tcg_out_ld(s, type, a0, a1, a2);
2595        break;
2596    case INDEX_op_st_vec:
2597        tcg_out_st(s, type, a0, a1, a2);
2598        break;
2599    case INDEX_op_add_vec:
2600        set_vtype_len_sew(s, type, vece);
2601        tcg_out_opc_vv_vi(s, OPC_VADD_VV, OPC_VADD_VI, a0, a1, a2, c2);
2602        break;
2603    case INDEX_op_sub_vec:
2604        set_vtype_len_sew(s, type, vece);
2605        if (const_args[1]) {
2606            tcg_out_opc_vi(s, OPC_VRSUB_VI, a0, a2, a1);
2607        } else {
2608            tcg_out_opc_vv(s, OPC_VSUB_VV, a0, a1, a2);
2609        }
2610        break;
2611    case INDEX_op_and_vec:
2612        set_vtype_len(s, type);
2613        tcg_out_opc_vv_vi(s, OPC_VAND_VV, OPC_VAND_VI, a0, a1, a2, c2);
2614        break;
2615    case INDEX_op_or_vec:
2616        set_vtype_len(s, type);
2617        tcg_out_opc_vv_vi(s, OPC_VOR_VV, OPC_VOR_VI, a0, a1, a2, c2);
2618        break;
2619    case INDEX_op_xor_vec:
2620        set_vtype_len(s, type);
2621        tcg_out_opc_vv_vi(s, OPC_VXOR_VV, OPC_VXOR_VI, a0, a1, a2, c2);
2622        break;
2623    case INDEX_op_not_vec:
2624        set_vtype_len(s, type);
2625        tcg_out_opc_vi(s, OPC_VXOR_VI, a0, a1, -1);
2626        break;
2627    case INDEX_op_neg_vec:
2628        set_vtype_len_sew(s, type, vece);
2629        tcg_out_opc_vi(s, OPC_VRSUB_VI, a0, a1, 0);
2630        break;
2631    case INDEX_op_mul_vec:
2632        set_vtype_len_sew(s, type, vece);
2633        tcg_out_opc_vv(s, OPC_VMUL_VV, a0, a1, a2);
2634        break;
2635    case INDEX_op_ssadd_vec:
2636        set_vtype_len_sew(s, type, vece);
2637        tcg_out_opc_vv_vi(s, OPC_VSADD_VV, OPC_VSADD_VI, a0, a1, a2, c2);
2638        break;
2639    case INDEX_op_sssub_vec:
2640        set_vtype_len_sew(s, type, vece);
2641        tcg_out_opc_vv_vi(s, OPC_VSSUB_VV, OPC_VSSUB_VI, a0, a1, a2, c2);
2642        break;
2643    case INDEX_op_usadd_vec:
2644        set_vtype_len_sew(s, type, vece);
2645        tcg_out_opc_vv_vi(s, OPC_VSADDU_VV, OPC_VSADDU_VI, a0, a1, a2, c2);
2646        break;
2647    case INDEX_op_ussub_vec:
2648        set_vtype_len_sew(s, type, vece);
2649        tcg_out_opc_vv_vi(s, OPC_VSSUBU_VV, OPC_VSSUBU_VI, a0, a1, a2, c2);
2650        break;
2651    case INDEX_op_smax_vec:
2652        set_vtype_len_sew(s, type, vece);
2653        tcg_out_opc_vv_vi(s, OPC_VMAX_VV, OPC_VMAX_VI, a0, a1, a2, c2);
2654        break;
2655    case INDEX_op_smin_vec:
2656        set_vtype_len_sew(s, type, vece);
2657        tcg_out_opc_vv_vi(s, OPC_VMIN_VV, OPC_VMIN_VI, a0, a1, a2, c2);
2658        break;
2659    case INDEX_op_umax_vec:
2660        set_vtype_len_sew(s, type, vece);
2661        tcg_out_opc_vv_vi(s, OPC_VMAXU_VV, OPC_VMAXU_VI, a0, a1, a2, c2);
2662        break;
2663    case INDEX_op_umin_vec:
2664        set_vtype_len_sew(s, type, vece);
2665        tcg_out_opc_vv_vi(s, OPC_VMINU_VV, OPC_VMINU_VI, a0, a1, a2, c2);
2666        break;
2667    case INDEX_op_shls_vec:
2668        set_vtype_len_sew(s, type, vece);
2669        tcg_out_opc_vx(s, OPC_VSLL_VX, a0, a1, a2);
2670        break;
2671    case INDEX_op_shrs_vec:
2672        set_vtype_len_sew(s, type, vece);
2673        tcg_out_opc_vx(s, OPC_VSRL_VX, a0, a1, a2);
2674        break;
2675    case INDEX_op_sars_vec:
2676        set_vtype_len_sew(s, type, vece);
2677        tcg_out_opc_vx(s, OPC_VSRA_VX, a0, a1, a2);
2678        break;
2679    case INDEX_op_shlv_vec:
2680        set_vtype_len_sew(s, type, vece);
2681        tcg_out_opc_vv(s, OPC_VSLL_VV, a0, a1, a2);
2682        break;
2683    case INDEX_op_shrv_vec:
2684        set_vtype_len_sew(s, type, vece);
2685        tcg_out_opc_vv(s, OPC_VSRL_VV, a0, a1, a2);
2686        break;
2687    case INDEX_op_sarv_vec:
2688        set_vtype_len_sew(s, type, vece);
2689        tcg_out_opc_vv(s, OPC_VSRA_VV, a0, a1, a2);
2690        break;
2691    case INDEX_op_shli_vec:
2692        set_vtype_len_sew(s, type, vece);
2693        tcg_out_vshifti(s, OPC_VSLL_VI, OPC_VSLL_VX, a0, a1, a2);
2694        break;
2695    case INDEX_op_shri_vec:
2696        set_vtype_len_sew(s, type, vece);
2697        tcg_out_vshifti(s, OPC_VSRL_VI, OPC_VSRL_VX, a0, a1, a2);
2698        break;
2699    case INDEX_op_sari_vec:
2700        set_vtype_len_sew(s, type, vece);
2701        tcg_out_vshifti(s, OPC_VSRA_VI, OPC_VSRA_VX, a0, a1, a2);
2702        break;
2703    case INDEX_op_rotli_vec:
2704        set_vtype_len_sew(s, type, vece);
2705        tcg_out_vshifti(s, OPC_VSLL_VI, OPC_VSLL_VX, TCG_REG_V0, a1, a2);
2706        tcg_out_vshifti(s, OPC_VSRL_VI, OPC_VSRL_VX, a0, a1,
2707                        -a2 & ((8 << vece) - 1));
2708        tcg_out_opc_vv(s, OPC_VOR_VV, a0, a0, TCG_REG_V0);
2709        break;
2710    case INDEX_op_rotls_vec:
2711        set_vtype_len_sew(s, type, vece);
2712        tcg_out_opc_vx(s, OPC_VSLL_VX, TCG_REG_V0, a1, a2);
2713        tcg_out_opc_reg(s, OPC_SUBW, TCG_REG_TMP0, TCG_REG_ZERO, a2);
2714        tcg_out_opc_vx(s, OPC_VSRL_VX, a0, a1, TCG_REG_TMP0);
2715        tcg_out_opc_vv(s, OPC_VOR_VV, a0, a0, TCG_REG_V0);
2716        break;
2717    case INDEX_op_rotlv_vec:
2718        set_vtype_len_sew(s, type, vece);
2719        tcg_out_opc_vi(s, OPC_VRSUB_VI, TCG_REG_V0, a2, 0);
2720        tcg_out_opc_vv(s, OPC_VSRL_VV, TCG_REG_V0, a1, TCG_REG_V0);
2721        tcg_out_opc_vv(s, OPC_VSLL_VV, a0, a1, a2);
2722        tcg_out_opc_vv(s, OPC_VOR_VV, a0, a0, TCG_REG_V0);
2723        break;
2724    case INDEX_op_rotrv_vec:
2725        set_vtype_len_sew(s, type, vece);
2726        tcg_out_opc_vi(s, OPC_VRSUB_VI, TCG_REG_V0, a2, 0);
2727        tcg_out_opc_vv(s, OPC_VSLL_VV, TCG_REG_V0, a1, TCG_REG_V0);
2728        tcg_out_opc_vv(s, OPC_VSRL_VV, a0, a1, a2);
2729        tcg_out_opc_vv(s, OPC_VOR_VV, a0, a0, TCG_REG_V0);
2730        break;
2731    case INDEX_op_cmp_vec:
2732        tcg_out_cmpsel(s, type, vece, args[3], a0, a1, a2, c2,
2733                       -1, true, 0, true);
2734        break;
2735    case INDEX_op_cmpsel_vec:
2736        tcg_out_cmpsel(s, type, vece, args[5], a0, a1, a2, c2,
2737                       args[3], const_args[3], args[4], const_args[4]);
2738        break;
2739    case INDEX_op_mov_vec: /* Always emitted via tcg_out_mov.  */
2740    case INDEX_op_dup_vec: /* Always emitted via tcg_out_dup_vec.  */
2741    default:
2742        g_assert_not_reached();
2743    }
2744}
2745
2746void tcg_expand_vec_op(TCGOpcode opc, TCGType type, unsigned vece,
2747                       TCGArg a0, ...)
2748{
2749    g_assert_not_reached();
2750}
2751
2752int tcg_can_emit_vec_op(TCGOpcode opc, TCGType type, unsigned vece)
2753{
2754    switch (opc) {
2755    case INDEX_op_add_vec:
2756    case INDEX_op_sub_vec:
2757    case INDEX_op_and_vec:
2758    case INDEX_op_or_vec:
2759    case INDEX_op_xor_vec:
2760    case INDEX_op_not_vec:
2761    case INDEX_op_neg_vec:
2762    case INDEX_op_mul_vec:
2763    case INDEX_op_ssadd_vec:
2764    case INDEX_op_sssub_vec:
2765    case INDEX_op_usadd_vec:
2766    case INDEX_op_ussub_vec:
2767    case INDEX_op_smax_vec:
2768    case INDEX_op_smin_vec:
2769    case INDEX_op_umax_vec:
2770    case INDEX_op_umin_vec:
2771    case INDEX_op_shls_vec:
2772    case INDEX_op_shrs_vec:
2773    case INDEX_op_sars_vec:
2774    case INDEX_op_shlv_vec:
2775    case INDEX_op_shrv_vec:
2776    case INDEX_op_sarv_vec:
2777    case INDEX_op_shri_vec:
2778    case INDEX_op_shli_vec:
2779    case INDEX_op_sari_vec:
2780    case INDEX_op_rotls_vec:
2781    case INDEX_op_rotlv_vec:
2782    case INDEX_op_rotrv_vec:
2783    case INDEX_op_rotli_vec:
2784    case INDEX_op_cmp_vec:
2785    case INDEX_op_cmpsel_vec:
2786        return 1;
2787    default:
2788        return 0;
2789    }
2790}
2791
2792static TCGConstraintSetIndex
2793tcg_target_op_def(TCGOpcode op, TCGType type, unsigned flags)
2794{
2795    switch (op) {
2796    case INDEX_op_goto_ptr:
2797        return C_O0_I1(r);
2798
2799    case INDEX_op_ld8u_i32:
2800    case INDEX_op_ld8s_i32:
2801    case INDEX_op_ld16u_i32:
2802    case INDEX_op_ld16s_i32:
2803    case INDEX_op_ld_i32:
2804    case INDEX_op_ld8u_i64:
2805    case INDEX_op_ld8s_i64:
2806    case INDEX_op_ld16u_i64:
2807    case INDEX_op_ld16s_i64:
2808    case INDEX_op_ld32s_i64:
2809    case INDEX_op_ld32u_i64:
2810    case INDEX_op_ld_i64:
2811        return C_O1_I1(r, r);
2812
2813    case INDEX_op_st8_i32:
2814    case INDEX_op_st16_i32:
2815    case INDEX_op_st_i32:
2816    case INDEX_op_st8_i64:
2817    case INDEX_op_st16_i64:
2818    case INDEX_op_st32_i64:
2819    case INDEX_op_st_i64:
2820        return C_O0_I2(rz, r);
2821
2822    case INDEX_op_qemu_ld_i32:
2823    case INDEX_op_qemu_ld_i64:
2824        return C_O1_I1(r, r);
2825    case INDEX_op_qemu_st_i32:
2826    case INDEX_op_qemu_st_i64:
2827        return C_O0_I2(rz, r);
2828
2829    case INDEX_op_st_vec:
2830        return C_O0_I2(v, r);
2831    case INDEX_op_dup_vec:
2832    case INDEX_op_dupm_vec:
2833    case INDEX_op_ld_vec:
2834        return C_O1_I1(v, r);
2835    case INDEX_op_neg_vec:
2836    case INDEX_op_not_vec:
2837    case INDEX_op_shli_vec:
2838    case INDEX_op_shri_vec:
2839    case INDEX_op_sari_vec:
2840    case INDEX_op_rotli_vec:
2841        return C_O1_I1(v, v);
2842    case INDEX_op_add_vec:
2843    case INDEX_op_and_vec:
2844    case INDEX_op_or_vec:
2845    case INDEX_op_xor_vec:
2846    case INDEX_op_ssadd_vec:
2847    case INDEX_op_sssub_vec:
2848    case INDEX_op_usadd_vec:
2849    case INDEX_op_ussub_vec:
2850    case INDEX_op_smax_vec:
2851    case INDEX_op_smin_vec:
2852    case INDEX_op_umax_vec:
2853    case INDEX_op_umin_vec:
2854        return C_O1_I2(v, v, vK);
2855    case INDEX_op_sub_vec:
2856        return C_O1_I2(v, vK, v);
2857    case INDEX_op_mul_vec:
2858    case INDEX_op_shlv_vec:
2859    case INDEX_op_shrv_vec:
2860    case INDEX_op_sarv_vec:
2861    case INDEX_op_rotlv_vec:
2862    case INDEX_op_rotrv_vec:
2863        return C_O1_I2(v, v, v);
2864    case INDEX_op_shls_vec:
2865    case INDEX_op_shrs_vec:
2866    case INDEX_op_sars_vec:
2867    case INDEX_op_rotls_vec:
2868        return C_O1_I2(v, v, r);
2869    case INDEX_op_cmp_vec:
2870        return C_O1_I2(v, v, vL);
2871    case INDEX_op_cmpsel_vec:
2872        return C_O1_I4(v, v, vL, vK, vK);
2873    default:
2874        return C_NotImplemented;
2875    }
2876}
2877
2878static const int tcg_target_callee_save_regs[] = {
2879    TCG_REG_S0,       /* used for the global env (TCG_AREG0) */
2880    TCG_REG_S1,
2881    TCG_REG_S2,
2882    TCG_REG_S3,
2883    TCG_REG_S4,
2884    TCG_REG_S5,
2885    TCG_REG_S6,
2886    TCG_REG_S7,
2887    TCG_REG_S8,
2888    TCG_REG_S9,
2889    TCG_REG_S10,
2890    TCG_REG_S11,
2891    TCG_REG_RA,       /* should be last for ABI compliance */
2892};
2893
2894/* Stack frame parameters.  */
2895#define REG_SIZE   (TCG_TARGET_REG_BITS / 8)
2896#define SAVE_SIZE  ((int)ARRAY_SIZE(tcg_target_callee_save_regs) * REG_SIZE)
2897#define TEMP_SIZE  (CPU_TEMP_BUF_NLONGS * (int)sizeof(long))
2898#define FRAME_SIZE ((TCG_STATIC_CALL_ARGS_SIZE + TEMP_SIZE + SAVE_SIZE \
2899                     + TCG_TARGET_STACK_ALIGN - 1) \
2900                    & -TCG_TARGET_STACK_ALIGN)
2901#define SAVE_OFS   (TCG_STATIC_CALL_ARGS_SIZE + TEMP_SIZE)
2902
2903/* We're expecting to be able to use an immediate for frame allocation.  */
2904QEMU_BUILD_BUG_ON(FRAME_SIZE > 0x7ff);
2905
2906/* Generate global QEMU prologue and epilogue code */
2907static void tcg_target_qemu_prologue(TCGContext *s)
2908{
2909    int i;
2910
2911    tcg_set_frame(s, TCG_REG_SP, TCG_STATIC_CALL_ARGS_SIZE, TEMP_SIZE);
2912
2913    /* TB prologue */
2914    tcg_out_opc_imm(s, OPC_ADDI, TCG_REG_SP, TCG_REG_SP, -FRAME_SIZE);
2915    for (i = 0; i < ARRAY_SIZE(tcg_target_callee_save_regs); i++) {
2916        tcg_out_st(s, TCG_TYPE_REG, tcg_target_callee_save_regs[i],
2917                   TCG_REG_SP, SAVE_OFS + i * REG_SIZE);
2918    }
2919
2920    if (!tcg_use_softmmu && guest_base) {
2921        tcg_out_movi(s, TCG_TYPE_PTR, TCG_GUEST_BASE_REG, guest_base);
2922        tcg_regset_set_reg(s->reserved_regs, TCG_GUEST_BASE_REG);
2923    }
2924
2925    /* Call generated code */
2926    tcg_out_mov(s, TCG_TYPE_PTR, TCG_AREG0, tcg_target_call_iarg_regs[0]);
2927    tcg_out_opc_imm(s, OPC_JALR, TCG_REG_ZERO, tcg_target_call_iarg_regs[1], 0);
2928
2929    /* Return path for goto_ptr. Set return value to 0 */
2930    tcg_code_gen_epilogue = tcg_splitwx_to_rx(s->code_ptr);
2931    tcg_out_mov(s, TCG_TYPE_REG, TCG_REG_A0, TCG_REG_ZERO);
2932
2933    /* TB epilogue */
2934    tb_ret_addr = tcg_splitwx_to_rx(s->code_ptr);
2935    for (i = 0; i < ARRAY_SIZE(tcg_target_callee_save_regs); i++) {
2936        tcg_out_ld(s, TCG_TYPE_REG, tcg_target_callee_save_regs[i],
2937                   TCG_REG_SP, SAVE_OFS + i * REG_SIZE);
2938    }
2939
2940    tcg_out_opc_imm(s, OPC_ADDI, TCG_REG_SP, TCG_REG_SP, FRAME_SIZE);
2941    tcg_out_opc_imm(s, OPC_JALR, TCG_REG_ZERO, TCG_REG_RA, 0);
2942}
2943
2944static void tcg_out_tb_start(TCGContext *s)
2945{
2946    init_setting_vtype(s);
2947}
2948
2949static bool vtype_check(unsigned vtype)
2950{
2951    unsigned long tmp;
2952
2953    /* vsetvl tmp, zero, vtype */
2954    asm(".insn r 0x57, 7, 0x40, %0, zero, %1" : "=r"(tmp) : "r"(vtype));
2955    return tmp != 0;
2956}
2957
2958static void probe_frac_lmul_1(TCGType type, MemOp vsew)
2959{
2960    VsetCache *p = &riscv_vset_cache[type - TCG_TYPE_V64][vsew];
2961    unsigned avl = tcg_type_size(type) >> vsew;
2962    int lmul = type - riscv_lg2_vlenb;
2963    unsigned vtype = encode_vtype(true, true, vsew, lmul & 7);
2964    bool lmul_eq_avl = true;
2965
2966    /* Guaranteed by Zve64x. */
2967    assert(lmul < 3);
2968
2969    /*
2970     * For LMUL < -3, the host vector size is so large that TYPE
2971     * is smaller than the minimum 1/8 fraction.
2972     *
2973     * For other fractional LMUL settings, implementations must
2974     * support SEW settings between SEW_MIN and LMUL * ELEN, inclusive.
2975     * So if ELEN = 64, LMUL = 1/2, then SEW will support e8, e16, e32,
2976     * but e64 may not be supported. In other words, the hardware only
2977     * guarantees SEW_MIN <= SEW <= LMUL * ELEN.  Check.
2978     */
2979    if (lmul < 0 && (lmul < -3 || !vtype_check(vtype))) {
2980        vtype = encode_vtype(true, true, vsew, VLMUL_M1);
2981        lmul_eq_avl = false;
2982    }
2983
2984    if (avl < 32) {
2985        p->vset_insn = encode_vseti(OPC_VSETIVLI, TCG_REG_ZERO, avl, vtype);
2986    } else if (lmul_eq_avl) {
2987        /* rd != 0 and rs1 == 0 uses vlmax */
2988        p->vset_insn = encode_vset(OPC_VSETVLI, TCG_REG_TMP0, TCG_REG_ZERO, vtype);
2989    } else {
2990        p->movi_insn = encode_i(OPC_ADDI, TCG_REG_TMP0, TCG_REG_ZERO, avl);
2991        p->vset_insn = encode_vset(OPC_VSETVLI, TCG_REG_ZERO, TCG_REG_TMP0, vtype);
2992    }
2993}
2994
2995static void probe_frac_lmul(void)
2996{
2997    /* Match riscv_lg2_vlenb to TCG_TYPE_V64. */
2998    QEMU_BUILD_BUG_ON(TCG_TYPE_V64 != 3);
2999
3000    for (TCGType t = TCG_TYPE_V64; t <= TCG_TYPE_V256; t++) {
3001        for (MemOp e = MO_8; e <= MO_64; e++) {
3002            probe_frac_lmul_1(t, e);
3003        }
3004    }
3005}
3006
3007static void tcg_target_init(TCGContext *s)
3008{
3009    tcg_target_available_regs[TCG_TYPE_I32] = 0xffffffff;
3010    tcg_target_available_regs[TCG_TYPE_I64] = 0xffffffff;
3011
3012    tcg_target_call_clobber_regs = -1;
3013    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_S0);
3014    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_S1);
3015    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_S2);
3016    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_S3);
3017    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_S4);
3018    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_S5);
3019    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_S6);
3020    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_S7);
3021    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_S8);
3022    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_S9);
3023    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_S10);
3024    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_S11);
3025
3026    s->reserved_regs = 0;
3027    tcg_regset_set_reg(s->reserved_regs, TCG_REG_ZERO);
3028    tcg_regset_set_reg(s->reserved_regs, TCG_REG_TMP0);
3029    tcg_regset_set_reg(s->reserved_regs, TCG_REG_TMP1);
3030    tcg_regset_set_reg(s->reserved_regs, TCG_REG_TMP2);
3031    tcg_regset_set_reg(s->reserved_regs, TCG_REG_SP);
3032    tcg_regset_set_reg(s->reserved_regs, TCG_REG_GP);
3033    tcg_regset_set_reg(s->reserved_regs, TCG_REG_TP);
3034
3035    if (cpuinfo & CPUINFO_ZVE64X) {
3036        switch (riscv_lg2_vlenb) {
3037        case TCG_TYPE_V64:
3038            tcg_target_available_regs[TCG_TYPE_V64] = ALL_VECTOR_REGS;
3039            tcg_target_available_regs[TCG_TYPE_V128] = ALL_DVECTOR_REG_GROUPS;
3040            tcg_target_available_regs[TCG_TYPE_V256] = ALL_QVECTOR_REG_GROUPS;
3041            s->reserved_regs |= (~ALL_QVECTOR_REG_GROUPS & ALL_VECTOR_REGS);
3042            break;
3043        case TCG_TYPE_V128:
3044            tcg_target_available_regs[TCG_TYPE_V64] = ALL_VECTOR_REGS;
3045            tcg_target_available_regs[TCG_TYPE_V128] = ALL_VECTOR_REGS;
3046            tcg_target_available_regs[TCG_TYPE_V256] = ALL_DVECTOR_REG_GROUPS;
3047            s->reserved_regs |= (~ALL_DVECTOR_REG_GROUPS & ALL_VECTOR_REGS);
3048            break;
3049        default:
3050            /* Guaranteed by Zve64x. */
3051            tcg_debug_assert(riscv_lg2_vlenb >= TCG_TYPE_V256);
3052            tcg_target_available_regs[TCG_TYPE_V64] = ALL_VECTOR_REGS;
3053            tcg_target_available_regs[TCG_TYPE_V128] = ALL_VECTOR_REGS;
3054            tcg_target_available_regs[TCG_TYPE_V256] = ALL_VECTOR_REGS;
3055            break;
3056        }
3057        tcg_regset_set_reg(s->reserved_regs, TCG_REG_V0);
3058        probe_frac_lmul();
3059    }
3060}
3061
3062typedef struct {
3063    DebugFrameHeader h;
3064    uint8_t fde_def_cfa[4];
3065    uint8_t fde_reg_ofs[ARRAY_SIZE(tcg_target_callee_save_regs) * 2];
3066} DebugFrame;
3067
3068#define ELF_HOST_MACHINE EM_RISCV
3069
3070static const DebugFrame debug_frame = {
3071    .h.cie.len = sizeof(DebugFrameCIE) - 4, /* length after .len member */
3072    .h.cie.id = -1,
3073    .h.cie.version = 1,
3074    .h.cie.code_align = 1,
3075    .h.cie.data_align = -(TCG_TARGET_REG_BITS / 8) & 0x7f, /* sleb128 */
3076    .h.cie.return_column = TCG_REG_RA,
3077
3078    /* Total FDE size does not include the "len" member.  */
3079    .h.fde.len = sizeof(DebugFrame) - offsetof(DebugFrame, h.fde.cie_offset),
3080
3081    .fde_def_cfa = {
3082        12, TCG_REG_SP,                 /* DW_CFA_def_cfa sp, ... */
3083        (FRAME_SIZE & 0x7f) | 0x80,     /* ... uleb128 FRAME_SIZE */
3084        (FRAME_SIZE >> 7)
3085    },
3086    .fde_reg_ofs = {
3087        0x80 + 9,  12,                  /* DW_CFA_offset, s1,  -96 */
3088        0x80 + 18, 11,                  /* DW_CFA_offset, s2,  -88 */
3089        0x80 + 19, 10,                  /* DW_CFA_offset, s3,  -80 */
3090        0x80 + 20, 9,                   /* DW_CFA_offset, s4,  -72 */
3091        0x80 + 21, 8,                   /* DW_CFA_offset, s5,  -64 */
3092        0x80 + 22, 7,                   /* DW_CFA_offset, s6,  -56 */
3093        0x80 + 23, 6,                   /* DW_CFA_offset, s7,  -48 */
3094        0x80 + 24, 5,                   /* DW_CFA_offset, s8,  -40 */
3095        0x80 + 25, 4,                   /* DW_CFA_offset, s9,  -32 */
3096        0x80 + 26, 3,                   /* DW_CFA_offset, s10, -24 */
3097        0x80 + 27, 2,                   /* DW_CFA_offset, s11, -16 */
3098        0x80 + 1 , 1,                   /* DW_CFA_offset, ra,  -8 */
3099    }
3100};
3101
3102void tcg_register_jit(const void *buf, size_t buf_size)
3103{
3104    tcg_register_jit_int(buf, buf_size, &debug_frame, sizeof(debug_frame));
3105}
3106