path: root/tcg/loongarch64/tcg-target.c.inc

/*
 * Tiny Code Generator for QEMU
 *
 * Copyright (c) 2021 WANG Xuerui <git@xen0n.name>
 *
 * Based on tcg/riscv/tcg-target.c.inc
 *
 * Copyright (c) 2018 SiFive, Inc
 * Copyright (c) 2008-2009 Arnaud Patard <arnaud.patard@rtp-net.org>
 * Copyright (c) 2009 Aurelien Jarno <aurelien@aurel32.net>
 * Copyright (c) 2008 Fabrice Bellard
 *
 * Permission is hereby granted, free of charge, to any person obtaining a copy
 * of this software and associated documentation files (the "Software"), to deal
 * in the Software without restriction, including without limitation the rights
 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
 * copies of the Software, and to permit persons to whom the Software is
 * furnished to do so, subject to the following conditions:
 *
 * The above copyright notice and this permission notice shall be included in
 * all copies or substantial portions of the Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
 * THE SOFTWARE.
 */

#include "../tcg-ldst.c.inc"

#ifdef CONFIG_DEBUG_TCG
static const char * const tcg_target_reg_names[TCG_TARGET_NB_REGS] = {
    "zero",
    "ra",
    "tp",
    "sp",
    "a0",
    "a1",
    "a2",
    "a3",
    "a4",
    "a5",
    "a6",
    "a7",
    "t0",
    "t1",
    "t2",
    "t3",
    "t4",
    "t5",
    "t6",
    "t7",
    "t8",
    "r21", /* reserved in the LP64* ABI, hence no ABI name */
    "s9",
    "s0",
    "s1",
    "s2",
    "s3",
    "s4",
    "s5",
    "s6",
    "s7",
    "s8"
};
#endif

static const int tcg_target_reg_alloc_order[] = {
    /* Registers preserved across calls */
    /* TCG_REG_S0 reserved for TCG_AREG0 */
    TCG_REG_S1,
    TCG_REG_S2,
    TCG_REG_S3,
    TCG_REG_S4,
    TCG_REG_S5,
    TCG_REG_S6,
    TCG_REG_S7,
    TCG_REG_S8,
    TCG_REG_S9,

    /* Registers (potentially) clobbered across calls */
    TCG_REG_T0,
    TCG_REG_T1,
    TCG_REG_T2,
    TCG_REG_T3,
    TCG_REG_T4,
    TCG_REG_T5,
    TCG_REG_T6,
    TCG_REG_T7,
    TCG_REG_T8,

    /* Argument registers, opposite order of allocation.  */
    TCG_REG_A7,
    TCG_REG_A6,
    TCG_REG_A5,
    TCG_REG_A4,
    TCG_REG_A3,
    TCG_REG_A2,
    TCG_REG_A1,
    TCG_REG_A0,
};

static const int tcg_target_call_iarg_regs[] = {
    TCG_REG_A0,
    TCG_REG_A1,
    TCG_REG_A2,
    TCG_REG_A3,
    TCG_REG_A4,
    TCG_REG_A5,
    TCG_REG_A6,
    TCG_REG_A7,
};

static const int tcg_target_call_oarg_regs[] = {
    TCG_REG_A0,
    TCG_REG_A1,
};

#ifndef CONFIG_SOFTMMU
#define USE_GUEST_BASE     (guest_base != 0)
#define TCG_GUEST_BASE_REG TCG_REG_S1
#endif

#define TCG_CT_CONST_ZERO  0x100
#define TCG_CT_CONST_S12   0x200
#define TCG_CT_CONST_S32   0x400
#define TCG_CT_CONST_U12   0x800
#define TCG_CT_CONST_C12   0x1000
#define TCG_CT_CONST_WSZ   0x2000

#define ALL_GENERAL_REGS      MAKE_64BIT_MASK(0, 32)
/*
 * For softmmu, we need to avoid conflicts with the first 5
 * argument registers to call the helper.  Some of these are
 * also used for the tlb lookup.
 */
#ifdef CONFIG_SOFTMMU
#define SOFTMMU_RESERVE_REGS  MAKE_64BIT_MASK(TCG_REG_A0, 5)
#else
#define SOFTMMU_RESERVE_REGS  0
#endif


static inline tcg_target_long sextreg(tcg_target_long val, int pos, int len)
{
    return sextract64(val, pos, len);
}

/* test if a constant matches the constraint */
static bool tcg_target_const_match(int64_t val, TCGType type, int ct)
{
    if (ct & TCG_CT_CONST) {
        return true;
    }
    if ((ct & TCG_CT_CONST_ZERO) && val == 0) {
        return true;
    }
    if ((ct & TCG_CT_CONST_S12) && val == sextreg(val, 0, 12)) {
        return true;
    }
    if ((ct & TCG_CT_CONST_S32) && val == (int32_t)val) {
        return true;
    }
    if ((ct & TCG_CT_CONST_U12) && val >= 0 && val <= 0xfff) {
        return true;
    }
    if ((ct & TCG_CT_CONST_C12) && ~val >= 0 && ~val <= 0xfff) {
        return true;
    }
    if ((ct & TCG_CT_CONST_WSZ) && val == (type == TCG_TYPE_I32 ? 32 : 64)) {
        return true;
    }
    return false;
}

/*
 * Relocations
 */

/*
 * Relocation records defined in LoongArch ELF psABI v1.00 is way too
 * complicated; a whopping stack machine is needed to stuff the fields, at
 * the very least one SOP_PUSH and one SOP_POP (of the correct format) are
 * needed.
 *
 * Hence, define our own simpler relocation types. Numbers are chosen as to
 * not collide with potential future additions to the true ELF relocation
 * type enum.
 */

/* Field Sk16, shifted right by 2; suitable for conditional jumps */
#define R_LOONGARCH_BR_SK16     256
/* Field Sd10k16, shifted right by 2; suitable for B and BL */
#define R_LOONGARCH_BR_SD10K16  257

static bool reloc_br_sk16(tcg_insn_unit *src_rw, const tcg_insn_unit *target)
{
    const tcg_insn_unit *src_rx = tcg_splitwx_to_rx(src_rw);
    intptr_t offset = (intptr_t)target - (intptr_t)src_rx;

    tcg_debug_assert((offset & 3) == 0);
    offset >>= 2;
    if (offset == sextreg(offset, 0, 16)) {
        *src_rw = deposit64(*src_rw, 10, 16, offset);
        return true;
    }

    return false;
}

static bool reloc_br_sd10k16(tcg_insn_unit *src_rw,
                             const tcg_insn_unit *target)
{
    const tcg_insn_unit *src_rx = tcg_splitwx_to_rx(src_rw);
    intptr_t offset = (intptr_t)target - (intptr_t)src_rx;

    tcg_debug_assert((offset & 3) == 0);
    offset >>= 2;
    if (offset == sextreg(offset, 0, 26)) {
        *src_rw = deposit64(*src_rw, 0, 10, offset >> 16); /* slot d10 */
        *src_rw = deposit64(*src_rw, 10, 16, offset); /* slot k16 */
        return true;
    }

    return false;
}

static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
                        intptr_t value, intptr_t addend)
{
    tcg_debug_assert(addend == 0);
    switch (type) {
    case R_LOONGARCH_BR_SK16:
        return reloc_br_sk16(code_ptr, (tcg_insn_unit *)value);
    case R_LOONGARCH_BR_SD10K16:
        return reloc_br_sd10k16(code_ptr, (tcg_insn_unit *)value);
    default:
        g_assert_not_reached();
    }
}

#include "tcg-insn-defs.c.inc"

/*
 * TCG intrinsics
 */

static void tcg_out_mb(TCGContext *s, TCGArg a0)
{
    /* Baseline LoongArch only has the full barrier, unfortunately.  */
    tcg_out_opc_dbar(s, 0);
}

static bool tcg_out_mov(TCGContext *s, TCGType type, TCGReg ret, TCGReg arg)
{
    if (ret == arg) {
        return true;
    }
    switch (type) {
    case TCG_TYPE_I32:
    case TCG_TYPE_I64:
        /*
         * Conventional register-register move used in LoongArch is
         * `or dst, src, zero`.
         */
        tcg_out_opc_or(s, ret, arg, TCG_REG_ZERO);
        break;
    default:
        g_assert_not_reached();
    }
    return true;
}

/* Loads a 32-bit immediate into rd, sign-extended.  */
static void tcg_out_movi_i32(TCGContext *s, TCGReg rd, int32_t val)
{
    tcg_target_long lo = sextreg(val, 0, 12);
    tcg_target_long hi12 = sextreg(val, 12, 20);

    /* Single-instruction cases.  */
    if (hi12 == 0) {
        /* val fits in uimm12: ori rd, zero, val */
        tcg_out_opc_ori(s, rd, TCG_REG_ZERO, val);
        return;
    }
    if (hi12 == sextreg(lo, 12, 20)) {
        /* val fits in simm12: addi.w rd, zero, val */
        tcg_out_opc_addi_w(s, rd, TCG_REG_ZERO, val);
        return;
    }

    /* High bits must be set; load with lu12i.w + optional ori.  */
    tcg_out_opc_lu12i_w(s, rd, hi12);
    if (lo != 0) {
        tcg_out_opc_ori(s, rd, rd, lo & 0xfff);
    }
}

static void tcg_out_movi(TCGContext *s, TCGType type, TCGReg rd,
                         tcg_target_long val)
{
    /*
     * LoongArch conventionally loads 64-bit immediates in at most 4 steps,
     * with dedicated instructions for filling the respective bitfields
     * below:
     *
     *        6                   5                   4               3
     *  3 2 1 0 9 8 7 6 5 4 3 2 1 0 9 8 7 6 5 4 3 2 1 0 9 8 7 6 5 4 3 2
     * +-----------------------+---------------------------------------+...
     * |          hi52         |                  hi32                 |
     * +-----------------------+---------------------------------------+...
     *       3                   2                   1
     *     1 0 9 8 7 6 5 4 3 2 1 0 9 8 7 6 5 4 3 2 1 0 9 8 7 6 5 4 3 2 1 0
     * ...+-------------------------------------+-------------------------+
     *    |                 hi12                |            lo           |
     * ...+-------------------------------------+-------------------------+
     *
     * Check if val belong to one of the several fast cases, before falling
     * back to the slow path.
     */

    intptr_t pc_offset;
    tcg_target_long val_lo, val_hi, pc_hi, offset_hi;
    tcg_target_long hi12, hi32, hi52;

    /* Value fits in signed i32.  */
    if (type == TCG_TYPE_I32 || val == (int32_t)val) {
        tcg_out_movi_i32(s, rd, val);
        return;
    }

    /* PC-relative cases.  */
    pc_offset = tcg_pcrel_diff(s, (void *)val);
    if (pc_offset == sextreg(pc_offset, 0, 22) && (pc_offset & 3) == 0) {
        /* Single pcaddu2i.  */
        tcg_out_opc_pcaddu2i(s, rd, pc_offset >> 2);
        return;
    }

    if (pc_offset == (int32_t)pc_offset) {
        /* Offset within 32 bits; load with pcalau12i + ori.  */
        val_lo = sextreg(val, 0, 12);
        val_hi = val >> 12;
        pc_hi = (val - pc_offset) >> 12;
        offset_hi = val_hi - pc_hi;

        tcg_debug_assert(offset_hi == sextreg(offset_hi, 0, 20));
        tcg_out_opc_pcalau12i(s, rd, offset_hi);
        if (val_lo != 0) {
            tcg_out_opc_ori(s, rd, rd, val_lo & 0xfff);
        }
        return;
    }

    hi12 = sextreg(val, 12, 20);
    hi32 = sextreg(val, 32, 20);
    hi52 = sextreg(val, 52, 12);

    /* Single cu52i.d case.  */
    if ((hi52 != 0) && (ctz64(val) >= 52)) {
        tcg_out_opc_cu52i_d(s, rd, TCG_REG_ZERO, hi52);
        return;
    }

    /* Slow path.  Initialize the low 32 bits, then concat high bits.  */
    tcg_out_movi_i32(s, rd, val);

    /* Load hi32 and hi52 explicitly when they are unexpected values. */
    if (hi32 != sextreg(hi12, 20, 20)) {
        tcg_out_opc_cu32i_d(s, rd, hi32);
    }

    if (hi52 != sextreg(hi32, 20, 12)) {
        tcg_out_opc_cu52i_d(s, rd, rd, hi52);
    }
}

static void tcg_out_addi(TCGContext *s, TCGType type, TCGReg rd,
                         TCGReg rs, tcg_target_long imm)
{
    tcg_target_long lo12 = sextreg(imm, 0, 12);
    tcg_target_long hi16 = sextreg(imm - lo12, 16, 16);

    /*
     * Note that there's a hole in between hi16 and lo12:
     *
     *       3                   2                   1                   0
     *     1 0 9 8 7 6 5 4 3 2 1 0 9 8 7 6 5 4 3 2 1 0 9 8 7 6 5 4 3 2 1 0
     * ...+-------------------------------+-------+-----------------------+
     *    |             hi16              |       |          lo12         |
     * ...+-------------------------------+-------+-----------------------+
     *
     * For bits within that hole, it's more efficient to use LU12I and ADD.
     */
    if (imm == (hi16 << 16) + lo12) {
        if (hi16) {
            tcg_out_opc_addu16i_d(s, rd, rs, hi16);
            rs = rd;
        }
        if (type == TCG_TYPE_I32) {
            tcg_out_opc_addi_w(s, rd, rs, lo12);
        } else if (lo12) {
            tcg_out_opc_addi_d(s, rd, rs, lo12);
        } else {
            tcg_out_mov(s, type, rd, rs);
        }
    } else {
        tcg_out_movi(s, type, TCG_REG_TMP0, imm);
        if (type == TCG_TYPE_I32) {
            tcg_out_opc_add_w(s, rd, rs, TCG_REG_TMP0);
        } else {
            tcg_out_opc_add_d(s, rd, rs, TCG_REG_TMP0);
        }
    }
}

static void tcg_out_ext8u(TCGContext *s, TCGReg ret, TCGReg arg)
{
    tcg_out_opc_andi(s, ret, arg, 0xff);
}

static void tcg_out_ext16u(TCGContext *s, TCGReg ret, TCGReg arg)
{
    tcg_out_opc_bstrpick_w(s, ret, arg, 0, 15);
}

static void tcg_out_ext32u(TCGContext *s, TCGReg ret, TCGReg arg)
{
    tcg_out_opc_bstrpick_d(s, ret, arg, 0, 31);
}

static void tcg_out_ext8s(TCGContext *s, TCGReg ret, TCGReg arg)
{
    tcg_out_opc_sext_b(s, ret, arg);
}

static void tcg_out_ext16s(TCGContext *s, TCGReg ret, TCGReg arg)
{
    tcg_out_opc_sext_h(s, ret, arg);
}

static void tcg_out_ext32s(TCGContext *s, TCGReg ret, TCGReg arg)
{
    tcg_out_opc_addi_w(s, ret, arg, 0);
}

static void tcg_out_clzctz(TCGContext *s, LoongArchInsn opc,
                           TCGReg a0, TCGReg a1, TCGReg a2,
                           bool c2, bool is_32bit)
{
    if (c2) {
        /*
         * Fast path: semantics already satisfied due to constraint and
         * insn behavior, single instruction is enough.
         */
        tcg_debug_assert(a2 == (is_32bit ? 32 : 64));
        /* all clz/ctz insns belong to DJ-format */
        tcg_out32(s, encode_dj_insn(opc, a0, a1));
        return;
    }

    tcg_out32(s, encode_dj_insn(opc, TCG_REG_TMP0, a1));
    /* a0 = a1 ? REG_TMP0 : a2 */
    tcg_out_opc_maskeqz(s, TCG_REG_TMP0, TCG_REG_TMP0, a1);
    tcg_out_opc_masknez(s, a0, a2, a1);
    tcg_out_opc_or(s, a0, TCG_REG_TMP0, a0);
}

static void tcg_out_setcond(TCGContext *s, TCGCond cond, TCGReg ret,
                            TCGReg arg1, TCGReg arg2, bool c2)
{
    TCGReg tmp;

    if (c2) {
        tcg_debug_assert(arg2 == 0);
    }

    switch (cond) {
    case TCG_COND_EQ:
        if (c2) {
            tmp = arg1;
        } else {
            tcg_out_opc_sub_d(s, ret, arg1, arg2);
            tmp = ret;
        }
        tcg_out_opc_sltui(s, ret, tmp, 1);
        break;
    case TCG_COND_NE:
        if (c2) {
            tmp = arg1;
        } else {
            tcg_out_opc_sub_d(s, ret, arg1, arg2);
            tmp = ret;
        }
        tcg_out_opc_sltu(s, ret, TCG_REG_ZERO, tmp);
        break;
    case TCG_COND_LT:
        tcg_out_opc_slt(s, ret, arg1, arg2);
        break;
    case TCG_COND_GE:
        tcg_out_opc_slt(s, ret, arg1, arg2);
        tcg_out_opc_xori(s, ret, ret, 1);
        break;
    case TCG_COND_LE:
        tcg_out_setcond(s, TCG_COND_GE, ret, arg2, arg1, false);
        break;
    case TCG_COND_GT:
        tcg_out_setcond(s, TCG_COND_LT, ret, arg2, arg1, false);
        break;
    case TCG_COND_LTU:
        tcg_out_opc_sltu(s, ret, arg1, arg2);
        break;
    case TCG_COND_GEU:
        tcg_out_opc_sltu(s, ret, arg1, arg2);
        tcg_out_opc_xori(s, ret, ret, 1);
        break;
    case TCG_COND_LEU:
        tcg_out_setcond(s, TCG_COND_GEU, ret, arg2, arg1, false);
        break;
    case TCG_COND_GTU:
        tcg_out_setcond(s, TCG_COND_LTU, ret, arg2, arg1, false);
        break;
    default:
        g_assert_not_reached();
        break;
    }
}

/*
 * Branch helpers
 */

static const struct {
    LoongArchInsn op;
    bool swap;
} tcg_brcond_to_loongarch[] = {
    [TCG_COND_EQ] =  { OPC_BEQ,  false },
    [TCG_COND_NE] =  { OPC_BNE,  false },
    [TCG_COND_LT] =  { OPC_BGT,  true  },
    [TCG_COND_GE] =  { OPC_BLE,  true  },
    [TCG_COND_LE] =  { OPC_BLE,  false },
    [TCG_COND_GT] =  { OPC_BGT,  false },
    [TCG_COND_LTU] = { OPC_BGTU, true  },
    [TCG_COND_GEU] = { OPC_BLEU, true  },
    [TCG_COND_LEU] = { OPC_BLEU, false },
    [TCG_COND_GTU] = { OPC_BGTU, false }
};

static void tcg_out_brcond(TCGContext *s, TCGCond cond, TCGReg arg1,
                           TCGReg arg2, TCGLabel *l)
{
    LoongArchInsn op = tcg_brcond_to_loongarch[cond].op;

    tcg_debug_assert(op != 0);

    if (tcg_brcond_to_loongarch[cond].swap) {
        TCGReg t = arg1;
        arg1 = arg2;
        arg2 = t;
    }

    /* all conditional branch insns belong to DJSk16-format */
    tcg_out_reloc(s, s->code_ptr, R_LOONGARCH_BR_SK16, l, 0);
    tcg_out32(s, encode_djsk16_insn(op, arg1, arg2, 0));
}

static void tcg_out_call_int(TCGContext *s, const tcg_insn_unit *arg, bool tail)
{
    TCGReg link = tail ? TCG_REG_ZERO : TCG_REG_RA;
    ptrdiff_t offset = tcg_pcrel_diff(s, arg);

    tcg_debug_assert((offset & 3) == 0);
    if (offset == sextreg(offset, 0, 28)) {
        /* short jump: +/- 256MiB */
        if (tail) {
            tcg_out_opc_b(s, offset >> 2);
        } else {
            tcg_out_opc_bl(s, offset >> 2);
        }
    } else if (offset == sextreg(offset, 0, 38)) {
        /* long jump: +/- 256GiB */
        tcg_target_long lo = sextreg(offset, 0, 18);
        tcg_target_long hi = offset - lo;
        tcg_out_opc_pcaddu18i(s, TCG_REG_TMP0, hi >> 18);
        tcg_out_opc_jirl(s, link, TCG_REG_TMP0, lo >> 2);
    } else {
        /* far jump: 64-bit */
        tcg_target_long lo = sextreg((tcg_target_long)arg, 0, 18);
        tcg_target_long hi = (tcg_target_long)arg - lo;
        tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_TMP0, hi);
        tcg_out_opc_jirl(s, link, TCG_REG_TMP0, lo >> 2);
    }
}

static void tcg_out_call(TCGContext *s, const tcg_insn_unit *arg,
                         const TCGHelperInfo *info)
{
    tcg_out_call_int(s, arg, false);
}

/*
 * Load/store helpers
 */

static void tcg_out_ldst(TCGContext *s, LoongArchInsn opc, TCGReg data,
                         TCGReg addr, intptr_t offset)
{
    intptr_t imm12 = sextreg(offset, 0, 12);

    if (offset != imm12) {
        intptr_t diff = offset - (uintptr_t)s->code_ptr;

        if (addr == TCG_REG_ZERO && diff == (int32_t)diff) {
            imm12 = sextreg(diff, 0, 12);
            tcg_out_opc_pcaddu12i(s, TCG_REG_TMP2, (diff - imm12) >> 12);
        } else {
            tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_TMP2, offset - imm12);
            if (addr != TCG_REG_ZERO) {
                tcg_out_opc_add_d(s, TCG_REG_TMP2, TCG_REG_TMP2, addr);
            }
        }
        addr = TCG_REG_TMP2;
    }

    switch (opc) {
    case OPC_LD_B:
    case OPC_LD_BU:
    case OPC_LD_H:
    case OPC_LD_HU:
    case OPC_LD_W:
    case OPC_LD_WU:
    case OPC_LD_D:
    case OPC_ST_B:
    case OPC_ST_H:
    case OPC_ST_W:
    case OPC_ST_D:
        tcg_out32(s, encode_djsk12_insn(opc, data, addr, imm12));
        break;
    default:
        g_assert_not_reached();
    }
}

static void tcg_out_ld(TCGContext *s, TCGType type, TCGReg arg,
                       TCGReg arg1, intptr_t arg2)
{
    bool is_32bit = type == TCG_TYPE_I32;
    tcg_out_ldst(s, is_32bit ? OPC_LD_W : OPC_LD_D, arg, arg1, arg2);
}

static void tcg_out_st(TCGContext *s, TCGType type, TCGReg arg,
                       TCGReg arg1, intptr_t arg2)
{
    bool is_32bit = type == TCG_TYPE_I32;
    tcg_out_ldst(s, is_32bit ? OPC_ST_W : OPC_ST_D, arg, arg1, arg2);
}

static bool tcg_out_sti(TCGContext *s, TCGType type, TCGArg val,
                        TCGReg base, intptr_t ofs)
{
    if (val == 0) {
        tcg_out_st(s, type, TCG_REG_ZERO, base, ofs);
        return true;
    }
    return false;
}

/*
 * Load/store helpers for SoftMMU, and qemu_ld/st implementations
 */

#if defined(CONFIG_SOFTMMU)
/*
 * helper signature: helper_ret_ld_mmu(CPUState *env, target_ulong addr,
 *                                     MemOpIdx oi, uintptr_t ra)
 */
static void * const qemu_ld_helpers[4] = {
    [MO_8]  = helper_ret_ldub_mmu,
    [MO_16] = helper_le_lduw_mmu,
    [MO_32] = helper_le_ldul_mmu,
    [MO_64] = helper_le_ldq_mmu,
};

/*
 * helper signature: helper_ret_st_mmu(CPUState *env, target_ulong addr,
 *                                     uintxx_t val, MemOpIdx oi,
 *                                     uintptr_t ra)
 */
static void * const qemu_st_helpers[4] = {
    [MO_8]  = helper_ret_stb_mmu,
    [MO_16] = helper_le_stw_mmu,
    [MO_32] = helper_le_stl_mmu,
    [MO_64] = helper_le_stq_mmu,
};

/* We expect to use a 12-bit negative offset from ENV.  */
QEMU_BUILD_BUG_ON(TLB_MASK_TABLE_OFS(0) > 0);
QEMU_BUILD_BUG_ON(TLB_MASK_TABLE_OFS(0) < -(1 << 11));

static bool tcg_out_goto(TCGContext *s, const tcg_insn_unit *target)
{
    tcg_out_opc_b(s, 0);
    return reloc_br_sd10k16(s->code_ptr - 1, target);
}

/*
 * Emits common code for TLB addend lookup, that eventually loads the
 * addend in TCG_REG_TMP2.
 */
static void tcg_out_tlb_load(TCGContext *s, TCGReg addrl, MemOpIdx oi,
                             tcg_insn_unit **label_ptr, bool is_load)
{
    MemOp opc = get_memop(oi);
    unsigned s_bits = opc & MO_SIZE;
    unsigned a_bits = get_alignment_bits(opc);
    tcg_target_long compare_mask;
    int mem_index = get_mmuidx(oi);
    int fast_ofs = TLB_MASK_TABLE_OFS(mem_index);
    int mask_ofs = fast_ofs + offsetof(CPUTLBDescFast, mask);
    int table_ofs = fast_ofs + offsetof(CPUTLBDescFast, table);

    tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_TMP0, TCG_AREG0, mask_ofs);
    tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_TMP1, TCG_AREG0, table_ofs);

    tcg_out_opc_srli_d(s, TCG_REG_TMP2, addrl,
                    TARGET_PAGE_BITS - CPU_TLB_ENTRY_BITS);
    tcg_out_opc_and(s, TCG_REG_TMP2, TCG_REG_TMP2, TCG_REG_TMP0);
    tcg_out_opc_add_d(s, TCG_REG_TMP2, TCG_REG_TMP2, TCG_REG_TMP1);

    /* Load the tlb comparator and the addend.  */
    tcg_out_ld(s, TCG_TYPE_TL, TCG_REG_TMP0, TCG_REG_TMP2,
               is_load ? offsetof(CPUTLBEntry, addr_read)
               : offsetof(CPUTLBEntry, addr_write));
    tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_TMP2, TCG_REG_TMP2,
               offsetof(CPUTLBEntry, addend));

    /* We don't support unaligned accesses.  */
    if (a_bits < s_bits) {
        a_bits = s_bits;
    }
    /* Clear the non-page, non-alignment bits from the address.  */
    compare_mask = (tcg_target_long)TARGET_PAGE_MASK | ((1 << a_bits) - 1);
    tcg_out_movi(s, TCG_TYPE_TL, TCG_REG_TMP1, compare_mask);
    tcg_out_opc_and(s, TCG_REG_TMP1, TCG_REG_TMP1, addrl);

    /* Compare masked address with the TLB entry.  */
    label_ptr[0] = s->code_ptr;
    tcg_out_opc_bne(s, TCG_REG_TMP0, TCG_REG_TMP1, 0);

    /* TLB Hit - addend in TCG_REG_TMP2, ready for use.  */
}

static void add_qemu_ldst_label(TCGContext *s, int is_ld, MemOpIdx oi,
                                TCGType type,
                                TCGReg datalo, TCGReg addrlo,
                                void *raddr, tcg_insn_unit **label_ptr)
{
    TCGLabelQemuLdst *label = new_ldst_label(s);

    label->is_ld = is_ld;
    label->oi = oi;
    label->type = type;
    label->datalo_reg = datalo;
    label->datahi_reg = 0; /* unused */
    label->addrlo_reg = addrlo;
    label->addrhi_reg = 0; /* unused */
    label->raddr = tcg_splitwx_to_rx(raddr);
    label->label_ptr[0] = label_ptr[0];
}

static bool tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *l)
{
    MemOpIdx oi = l->oi;
    MemOp opc = get_memop(oi);
    MemOp size = opc & MO_SIZE;
    TCGType type = l->type;

    /* resolve label address */
    if (!reloc_br_sk16(l->label_ptr[0], tcg_splitwx_to_rx(s->code_ptr))) {
        return false;
    }

    /* call load helper */
    tcg_out_mov(s, TCG_TYPE_PTR, TCG_REG_A0, TCG_AREG0);
    tcg_out_mov(s, TCG_TYPE_PTR, TCG_REG_A1, l->addrlo_reg);
    tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_A2, oi);
    tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_A3, (tcg_target_long)l->raddr);

    tcg_out_call_int(s, qemu_ld_helpers[size], false);

    switch (opc & MO_SSIZE) {
    case MO_SB:
        tcg_out_ext8s(s, l->datalo_reg, TCG_REG_A0);
        break;
    case MO_SW:
        tcg_out_ext16s(s, l->datalo_reg, TCG_REG_A0);
        break;
    case MO_SL:
        tcg_out_ext32s(s, l->datalo_reg, TCG_REG_A0);
        break;
    case MO_UL:
        if (type == TCG_TYPE_I32) {
            /* MO_UL loads of i32 should be sign-extended too */
            tcg_out_ext32s(s, l->datalo_reg, TCG_REG_A0);
            break;
        }
        /* fallthrough */
    default:
        tcg_out_mov(s, type, l->datalo_reg, TCG_REG_A0);
        break;
    }

    return tcg_out_goto(s, l->raddr);
}

static bool tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *l)
{
    MemOpIdx oi = l->oi;
    MemOp opc = get_memop(oi);
    MemOp size = opc & MO_SIZE;

    /* resolve label address */
    if (!reloc_br_sk16(l->label_ptr[0], tcg_splitwx_to_rx(s->code_ptr))) {
        return false;
    }

    /* call store helper */
    tcg_out_mov(s, TCG_TYPE_PTR, TCG_REG_A0, TCG_AREG0);
    tcg_out_mov(s, TCG_TYPE_PTR, TCG_REG_A1, l->addrlo_reg);
    switch (size) {
    case MO_8:
        tcg_out_ext8u(s, TCG_REG_A2, l->datalo_reg);
        break;
    case MO_16:
        tcg_out_ext16u(s, TCG_REG_A2, l->datalo_reg);
        break;
    case MO_32:
        tcg_out_ext32u(s, TCG_REG_A2, l->datalo_reg);
        break;
    case MO_64:
        tcg_out_mov(s, TCG_TYPE_I64, TCG_REG_A2, l->datalo_reg);
        break;
    default:
        g_assert_not_reached();
        break;
    }
    tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_A3, oi);
    tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_A4, (tcg_target_long)l->raddr);

    tcg_out_call_int(s, qemu_st_helpers[size], false);

    return tcg_out_goto(s, l->raddr);
}
#else

/*
 * Alignment helpers for user-mode emulation
 */

static void tcg_out_test_alignment(TCGContext *s, bool is_ld, TCGReg addr_reg,
                                   unsigned a_bits)
{
    TCGLabelQemuLdst *l = new_ldst_label(s);

    l->is_ld = is_ld;
    l->addrlo_reg = addr_reg;

    /*
     * Without micro-architecture details, we don't know which of bstrpick or
     * andi is faster, so use bstrpick as it's not constrained by imm field
     * width. (Not to say alignments >= 2^12 are going to happen any time
     * soon, though)
     */
    tcg_out_opc_bstrpick_d(s, TCG_REG_TMP1, addr_reg, 0, a_bits - 1);

    l->label_ptr[0] = s->code_ptr;
    tcg_out_opc_bne(s, TCG_REG_TMP1, TCG_REG_ZERO, 0);

    l->raddr = tcg_splitwx_to_rx(s->code_ptr);
}

static bool tcg_out_fail_alignment(TCGContext *s, TCGLabelQemuLdst *l)
{
    /* resolve label address */
    if (!reloc_br_sk16(l->label_ptr[0], tcg_splitwx_to_rx(s->code_ptr))) {
        return false;
    }

    tcg_out_mov(s, TCG_TYPE_TL, TCG_REG_A1, l->addrlo_reg);
    tcg_out_mov(s, TCG_TYPE_PTR, TCG_REG_A0, TCG_AREG0);

    /* tail call, with the return address back inline. */
    tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_RA, (uintptr_t)l->raddr);
    tcg_out_call_int(s, (const void *)(l->is_ld ? helper_unaligned_ld
                                       : helper_unaligned_st), true);
    return true;
}

static bool tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *l)
{
    return tcg_out_fail_alignment(s, l);
}

static bool tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *l)
{
    return tcg_out_fail_alignment(s, l);
}

#endif /* CONFIG_SOFTMMU */

/*
 * `ext32u` the address register into the temp register given,
 * if target is 32-bit, no-op otherwise.
 *
 * Returns the address register ready for use with TLB addend.
 */
static TCGReg tcg_out_zext_addr_if_32_bit(TCGContext *s,
                                          TCGReg addr, TCGReg tmp)
{
    if (TARGET_LONG_BITS == 32) {
        tcg_out_ext32u(s, tmp, addr);
        return tmp;
    }
    return addr;
}

static void tcg_out_qemu_ld_indexed(TCGContext *s, TCGReg rd, TCGReg rj,
                                   TCGReg rk, MemOp opc, TCGType type)
{
    /* Byte swapping is left to middle-end expansion.  */
    tcg_debug_assert((opc & MO_BSWAP) == 0);

    switch (opc & MO_SSIZE) {
    case MO_UB:
        tcg_out_opc_ldx_bu(s, rd, rj, rk);
        break;
    case MO_SB:
        tcg_out_opc_ldx_b(s, rd, rj, rk);
        break;
    case MO_UW:
        tcg_out_opc_ldx_hu(s, rd, rj, rk);
        break;
    case MO_SW:
        tcg_out_opc_ldx_h(s, rd, rj, rk);
        break;
    case MO_UL:
        if (type == TCG_TYPE_I64) {
            tcg_out_opc_ldx_wu(s, rd, rj, rk);
            break;
        }
        /* fallthrough */
    case MO_SL:
        tcg_out_opc_ldx_w(s, rd, rj, rk);
        break;
    case MO_UQ:
        tcg_out_opc_ldx_d(s, rd, rj, rk);
        break;
    default:
        g_assert_not_reached();
    }
}

static void tcg_out_qemu_ld(TCGContext *s, const TCGArg *args, TCGType type)
{
    TCGReg addr_regl;
    TCGReg data_regl;
    MemOpIdx oi;
    MemOp opc;
#if defined(CONFIG_SOFTMMU)
    tcg_insn_unit *label_ptr[1];
#else
    unsigned a_bits;
#endif
    TCGReg base;

    data_regl = *args++;
    addr_regl = *args++;
    oi = *args++;
    opc = get_memop(oi);

#if defined(CONFIG_SOFTMMU)
    tcg_out_tlb_load(s, addr_regl, oi, label_ptr, 1);
    base = tcg_out_zext_addr_if_32_bit(s, addr_regl, TCG_REG_TMP0);
    tcg_out_qemu_ld_indexed(s, data_regl, base, TCG_REG_TMP2, opc, type);
    add_qemu_ldst_label(s, 1, oi, type,
                        data_regl, addr_regl,
                        s->code_ptr, label_ptr);
#else
    a_bits = get_alignment_bits(opc);
    if (a_bits) {
        tcg_out_test_alignment(s, true, addr_regl, a_bits);
    }
    base = tcg_out_zext_addr_if_32_bit(s, addr_regl, TCG_REG_TMP0);
    TCGReg guest_base_reg = USE_GUEST_BASE ? TCG_GUEST_BASE_REG : TCG_REG_ZERO;
    tcg_out_qemu_ld_indexed(s, data_regl, base, guest_base_reg, opc, type);
#endif
}

static void tcg_out_qemu_st_indexed(TCGContext *s, TCGReg data,
                                   TCGReg rj, TCGReg rk, MemOp opc)
{
    /* Byte swapping is left to middle-end expansion.  */
    tcg_debug_assert((opc & MO_BSWAP) == 0);

    switch (opc & MO_SIZE) {
    case MO_8:
        tcg_out_opc_stx_b(s, data, rj, rk);
        break;
    case MO_16:
        tcg_out_opc_stx_h(s, data, rj, rk);
        break;
    case MO_32:
        tcg_out_opc_stx_w(s, data, rj, rk);
        break;
    case MO_64:
        tcg_out_opc_stx_d(s, data, rj, rk);
        break;
    default:
        g_assert_not_reached();
    }
}

static void tcg_out_qemu_st(TCGContext *s, const TCGArg *args)
{
    TCGReg addr_regl;
    TCGReg data_regl;
    MemOpIdx oi;
    MemOp opc;
#if defined(CONFIG_SOFTMMU)
    tcg_insn_unit *label_ptr[1];
#else
    unsigned a_bits;
#endif
    TCGReg base;

    data_regl = *args++;
    addr_regl = *args++;
    oi = *args++;
    opc = get_memop(oi);

#if defined(CONFIG_SOFTMMU)
    tcg_out_tlb_load(s, addr_regl, oi, label_ptr, 0);
    base = tcg_out_zext_addr_if_32_bit(s, addr_regl, TCG_REG_TMP0);
    tcg_out_qemu_st_indexed(s, data_regl, base, TCG_REG_TMP2, opc);
    add_qemu_ldst_label(s, 0, oi,
                        0, /* type param is unused for stores */
                        data_regl, addr_regl,
                        s->code_ptr, label_ptr);
#else
    a_bits = get_alignment_bits(opc);
    if (a_bits) {
        tcg_out_test_alignment(s, false, addr_regl, a_bits);
    }
    base = tcg_out_zext_addr_if_32_bit(s, addr_regl, TCG_REG_TMP0);
    TCGReg guest_base_reg = USE_GUEST_BASE ? TCG_GUEST_BASE_REG : TCG_REG_ZERO;
    tcg_out_qemu_st_indexed(s, data_regl, base, guest_base_reg, opc);
#endif
}

/* LoongArch uses `andi zero, zero, 0` as NOP.  */
#define NOP OPC_ANDI
static void tcg_out_nop(TCGContext *s)
{
    tcg_out32(s, NOP);
}

void tb_target_set_jmp_target(const TranslationBlock *tb, int n,
                              uintptr_t jmp_rx, uintptr_t jmp_rw)
{
    tcg_insn_unit i1, i2;
    ptrdiff_t upper, lower;
    uintptr_t addr = tb->jmp_target_addr[n];
    ptrdiff_t offset = (ptrdiff_t)(addr - jmp_rx) >> 2;

    if (offset == sextreg(offset, 0, 26)) {
        i1 = encode_sd10k16_insn(OPC_B, offset);
        i2 = NOP;
    } else {
        tcg_debug_assert(offset == sextreg(offset, 0, 36));
        lower = (int16_t)offset;
        upper = (offset - lower) >> 16;

        i1 = encode_dsj20_insn(OPC_PCADDU18I, TCG_REG_TMP0, upper);
        i2 = encode_djsk16_insn(OPC_JIRL, TCG_REG_ZERO, TCG_REG_TMP0, lower);
    }
    uint64_t pair = ((uint64_t)i2 << 32) | i1;
    qatomic_set((uint64_t *)jmp_rw, pair);
    flush_idcache_range(jmp_rx, jmp_rw, 8);
}

/*
 * Entry-points
 */

static const tcg_insn_unit *tb_ret_addr;

static void tcg_out_exit_tb(TCGContext *s, uintptr_t a0)
{
    /* Reuse the zeroing that exists for goto_ptr.  */
    if (a0 == 0) {
        tcg_out_call_int(s, tcg_code_gen_epilogue, true);
    } else {
        tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_A0, a0);
        tcg_out_call_int(s, tb_ret_addr, true);
    }
}

static void tcg_out_goto_tb(TCGContext *s, int which)
{
    /*
     * Ensure that patch area is 8-byte aligned so that an
     * atomic write can be used to patch the target address.
     */
    if ((uintptr_t)s->code_ptr & 7) {
        tcg_out_nop(s);
    }
    set_jmp_insn_offset(s, which);
    /*
     * actual branch destination will be patched by
     * tb_target_set_jmp_target later
     */
    tcg_out_opc_pcaddu18i(s, TCG_REG_TMP0, 0);
    tcg_out_opc_jirl(s, TCG_REG_ZERO, TCG_REG_TMP0, 0);
    set_jmp_reset_offset(s, which);
}

static void tcg_out_op(TCGContext *s, TCGOpcode opc,
                       const TCGArg args[TCG_MAX_OP_ARGS],
                       const int const_args[TCG_MAX_OP_ARGS])
{
    TCGArg a0 = args[0];
    TCGArg a1 = args[1];
    TCGArg a2 = args[2];
    int c2 = const_args[2];

    switch (opc) {
    case INDEX_op_mb:
        tcg_out_mb(s, a0);
        break;

    case INDEX_op_goto_ptr:
        tcg_out_opc_jirl(s, TCG_REG_ZERO, a0, 0);
        break;

    case INDEX_op_br:
        tcg_out_reloc(s, s->code_ptr, R_LOONGARCH_BR_SD10K16, arg_label(a0),
                      0);
        tcg_out_opc_b(s, 0);
        break;

    case INDEX_op_brcond_i32:
    case INDEX_op_brcond_i64:
        tcg_out_brcond(s, a2, a0, a1, arg_label(args[3]));
        break;

    case INDEX_op_ext8s_i32:
    case INDEX_op_ext8s_i64:
        tcg_out_ext8s(s, a0, a1);
        break;

    case INDEX_op_ext8u_i32:
    case INDEX_op_ext8u_i64:
        tcg_out_ext8u(s, a0, a1);
        break;

    case INDEX_op_ext16s_i32:
    case INDEX_op_ext16s_i64:
        tcg_out_ext16s(s, a0, a1);
        break;

    case INDEX_op_ext16u_i32:
    case INDEX_op_ext16u_i64:
        tcg_out_ext16u(s, a0, a1);
        break;

    case INDEX_op_ext32u_i64:
    case INDEX_op_extu_i32_i64:
        tcg_out_ext32u(s, a0, a1);
        break;

    case INDEX_op_ext32s_i64:
    case INDEX_op_extrl_i64_i32:
    case INDEX_op_ext_i32_i64:
        tcg_out_ext32s(s, a0, a1);
        break;

    case INDEX_op_extrh_i64_i32:
        tcg_out_opc_srai_d(s, a0, a1, 32);
        break;

    case INDEX_op_not_i32:
    case INDEX_op_not_i64:
        tcg_out_opc_nor(s, a0, a1, TCG_REG_ZERO);
        break;

    case INDEX_op_nor_i32:
    case INDEX_op_nor_i64:
        if (c2) {
            tcg_out_opc_ori(s, a0, a1, a2);
            tcg_out_opc_nor(s, a0, a0, TCG_REG_ZERO);
        } else {
            tcg_out_opc_nor(s, a0, a1, a2);
        }
        break;

    case INDEX_op_andc_i32:
    case INDEX_op_andc_i64:
        if (c2) {
            /* guaranteed to fit due to constraint */
            tcg_out_opc_andi(s, a0, a1, ~a2);
        } else {
            tcg_out_opc_andn(s, a0, a1, a2);
        }
        break;

    case INDEX_op_orc_i32:
    case INDEX_op_orc_i64:
        if (c2) {
            /* guaranteed to fit due to constraint */
            tcg_out_opc_ori(s, a0, a1, ~a2);
        } else {
            tcg_out_opc_orn(s, a0, a1, a2);
        }
        break;

    case INDEX_op_and_i32:
    case INDEX_op_and_i64:
        if (c2) {
            tcg_out_opc_andi(s, a0, a1, a2);
        } else {
            tcg_out_opc_and(s, a0, a1, a2);
        }
        break;

    case INDEX_op_or_i32:
    case INDEX_op_or_i64:
        if (c2) {
            tcg_out_opc_ori(s, a0, a1, a2);
        } else {
            tcg_out_opc_or(s, a0, a1, a2);
        }
        break;

    case INDEX_op_xor_i32:
    case INDEX_op_xor_i64:
        if (c2) {
            tcg_out_opc_xori(s, a0, a1, a2);
        } else {
            tcg_out_opc_xor(s, a0, a1, a2);
        }
        break;

    case INDEX_op_extract_i32:
        tcg_out_opc_bstrpick_w(s, a0, a1, a2, a2 + args[3] - 1);
        break;
    case INDEX_op_extract_i64:
        tcg_out_opc_bstrpick_d(s, a0, a1, a2, a2 + args[3] - 1);
        break;

    case INDEX_op_deposit_i32:
        tcg_out_opc_bstrins_w(s, a0, a2, args[3], args[3] + args[4] - 1);
        break;
    case INDEX_op_deposit_i64:
        tcg_out_opc_bstrins_d(s, a0, a2, args[3], args[3] + args[4] - 1);
        break;

    case INDEX_op_bswap16_i32:
    case INDEX_op_bswap16_i64:
        tcg_out_opc_revb_2h(s, a0, a1);
        if (a2 & TCG_BSWAP_OS) {
            tcg_out_ext16s(s, a0, a0);
        } else if ((a2 & (TCG_BSWAP_IZ | TCG_BSWAP_OZ)) == TCG_BSWAP_OZ) {
            tcg_out_ext16u(s, a0, a0);
        }
        break;

    case INDEX_op_bswap32_i32:
        /* All 32-bit values are computed sign-extended in the register.  */
        a2 = TCG_BSWAP_OS;
        /* fallthrough */
    case INDEX_op_bswap32_i64:
        tcg_out_opc_revb_2w(s, a0, a1);
        if (a2 & TCG_BSWAP_OS) {
            tcg_out_ext32s(s, a0, a0);
        } else if ((a2 & (TCG_BSWAP_IZ | TCG_BSWAP_OZ)) == TCG_BSWAP_OZ) {
            tcg_out_ext32u(s, a0, a0);
        }
        break;

    case INDEX_op_bswap64_i64:
        tcg_out_opc_revb_d(s, a0, a1);
        break;

    case INDEX_op_clz_i32:
        tcg_out_clzctz(s, OPC_CLZ_W, a0, a1, a2, c2, true);
        break;
    case INDEX_op_clz_i64:
        tcg_out_clzctz(s, OPC_CLZ_D, a0, a1, a2, c2, false);
        break;

    case INDEX_op_ctz_i32:
        tcg_out_clzctz(s, OPC_CTZ_W, a0, a1, a2, c2, true);
        break;
    case INDEX_op_ctz_i64:
        tcg_out_clzctz(s, OPC_CTZ_D, a0, a1, a2, c2, false);
        break;

    case INDEX_op_shl_i32:
        if (c2) {
            tcg_out_opc_slli_w(s, a0, a1, a2 & 0x1f);
        } else {
            tcg_out_opc_sll_w(s, a0, a1, a2);
        }
        break;
    case INDEX_op_shl_i64:
        if (c2) {
            tcg_out_opc_slli_d(s, a0, a1, a2 & 0x3f);
        } else {
            tcg_out_opc_sll_d(s, a0, a1, a2);
        }
        break;

    case INDEX_op_shr_i32:
        if (c2) {
            tcg_out_opc_srli_w(s, a0, a1, a2 & 0x1f);
        } else {
            tcg_out_opc_srl_w(s, a0, a1, a2);
        }
        break;
    case INDEX_op_shr_i64:
        if (c2) {
            tcg_out_opc_srli_d(s, a0, a1, a2 & 0x3f);
        } else {
            tcg_out_opc_srl_d(s, a0, a1, a2);
        }
        break;

    case INDEX_op_sar_i32:
        if (c2) {
            tcg_out_opc_srai_w(s, a0, a1, a2 & 0x1f);
        } else {
            tcg_out_opc_sra_w(s, a0, a1, a2);
        }
        break;
    case INDEX_op_sar_i64:
        if (c2) {
            tcg_out_opc_srai_d(s, a0, a1, a2 & 0x3f);
        } else {
            tcg_out_opc_sra_d(s, a0, a1, a2);
        }
        break;

    case INDEX_op_rotl_i32:
        /* transform into equivalent rotr/rotri */
        if (c2) {
            tcg_out_opc_rotri_w(s, a0, a1, (32 - a2) & 0x1f);
        } else {
            tcg_out_opc_sub_w(s, TCG_REG_TMP0, TCG_REG_ZERO, a2);
            tcg_out_opc_rotr_w(s, a0, a1, TCG_REG_TMP0);
        }
        break;
    case INDEX_op_rotl_i64:
        /* transform into equivalent rotr/rotri */
        if (c2) {
            tcg_out_opc_rotri_d(s, a0, a1, (64 - a2) & 0x3f);
        } else {
            tcg_out_opc_sub_w(s, TCG_REG_TMP0, TCG_REG_ZERO, a2);
            tcg_out_opc_rotr_d(s, a0, a1, TCG_REG_TMP0);
        }
        break;

    case INDEX_op_rotr_i32:
        if (c2) {
            tcg_out_opc_rotri_w(s, a0, a1, a2 & 0x1f);
        } else {
            tcg_out_opc_rotr_w(s, a0, a1, a2);
        }
        break;
    case INDEX_op_rotr_i64:
        if (c2) {
            tcg_out_opc_rotri_d(s, a0, a1, a2 & 0x3f);
        } else {
            tcg_out_opc_rotr_d(s, a0, a1, a2);
        }
        break;

    case INDEX_op_add_i32:
        if (c2) {
            tcg_out_addi(s, TCG_TYPE_I32, a0, a1, a2);
        } else {
            tcg_out_opc_add_w(s, a0, a1, a2);
        }
        break;
    case INDEX_op_add_i64:
        if (c2) {
            tcg_out_addi(s, TCG_TYPE_I64, a0, a1, a2);
        } else {
            tcg_out_opc_add_d(s, a0, a1, a2);
        }
        break;

    case INDEX_op_sub_i32:
        if (c2) {
            tcg_out_addi(s, TCG_TYPE_I32, a0, a1, -a2);
        } else {
            tcg_out_opc_sub_w(s, a0, a1, a2);
        }
        break;
    case INDEX_op_sub_i64:
        if (c2) {
            tcg_out_addi(s, TCG_TYPE_I64, a0, a1, -a2);
        } else {
            tcg_out_opc_sub_d(s, a0, a1, a2);
        }
        break;

    case INDEX_op_mul_i32:
        tcg_out_opc_mul_w(s, a0, a1, a2);
        break;
    case INDEX_op_mul_i64:
        tcg_out_opc_mul_d(s, a0, a1, a2);
        break;

    case INDEX_op_mulsh_i32:
        tcg_out_opc_mulh_w(s, a0, a1, a2);
        break;
    case INDEX_op_mulsh_i64:
        tcg_out_opc_mulh_d(s, a0, a1, a2);
        break;

    case INDEX_op_muluh_i32:
        tcg_out_opc_mulh_wu(s, a0, a1, a2);
        break;
    case INDEX_op_muluh_i64:
        tcg_out_opc_mulh_du(s, a0, a1, a2);
        break;

    case INDEX_op_div_i32:
        tcg_out_opc_div_w(s, a0, a1, a2);
        break;
    case INDEX_op_div_i64:
        tcg_out_opc_div_d(s, a0, a1, a2);
        break;

    case INDEX_op_divu_i32:
        tcg_out_opc_div_wu(s, a0, a1, a2);
        break;
    case INDEX_op_divu_i64:
        tcg_out_opc_div_du(s, a0, a1, a2);
        break;

    case INDEX_op_rem_i32:
        tcg_out_opc_mod_w(s, a0, a1, a2);
        break;
    case INDEX_op_rem_i64:
        tcg_out_opc_mod_d(s, a0, a1, a2);
        break;

    case INDEX_op_remu_i32:
        tcg_out_opc_mod_wu(s, a0, a1, a2);
        break;
    case INDEX_op_remu_i64:
        tcg_out_opc_mod_du(s, a0, a1, a2);
        break;

    case INDEX_op_setcond_i32:
    case INDEX_op_setcond_i64:
        tcg_out_setcond(s, args[3], a0, a1, a2, c2);
        break;

    case INDEX_op_ld8s_i32:
    case INDEX_op_ld8s_i64:
        tcg_out_ldst(s, OPC_LD_B, a0, a1, a2);
        break;
    case INDEX_op_ld8u_i32:
    case INDEX_op_ld8u_i64:
        tcg_out_ldst(s, OPC_LD_BU, a0, a1, a2);
        break;
    case INDEX_op_ld16s_i32:
    case INDEX_op_ld16s_i64:
        tcg_out_ldst(s, OPC_LD_H, a0, a1, a2);
        break;
    case INDEX_op_ld16u_i32:
    case INDEX_op_ld16u_i64:
        tcg_out_ldst(s, OPC_LD_HU, a0, a1, a2);
        break;
    case INDEX_op_ld_i32:
    case INDEX_op_ld32s_i64:
        tcg_out_ldst(s, OPC_LD_W, a0, a1, a2);
        break;
    case INDEX_op_ld32u_i64:
        tcg_out_ldst(s, OPC_LD_WU, a0, a1, a2);
        break;
    case INDEX_op_ld_i64:
        tcg_out_ldst(s, OPC_LD_D, a0, a1, a2);
        break;

    case INDEX_op_st8_i32:
    case INDEX_op_st8_i64:
        tcg_out_ldst(s, OPC_ST_B, a0, a1, a2);
        break;
    case INDEX_op_st16_i32:
    case INDEX_op_st16_i64:
        tcg_out_ldst(s, OPC_ST_H, a0, a1, a2);
        break;
    case INDEX_op_st_i32:
    case INDEX_op_st32_i64:
        tcg_out_ldst(s, OPC_ST_W, a0, a1, a2);
        break;
    case INDEX_op_st_i64:
        tcg_out_ldst(s, OPC_ST_D, a0, a1, a2);
        break;

    case INDEX_op_qemu_ld_i32:
        tcg_out_qemu_ld(s, args, TCG_TYPE_I32);
        break;
    case INDEX_op_qemu_ld_i64:
        tcg_out_qemu_ld(s, args, TCG_TYPE_I64);
        break;
    case INDEX_op_qemu_st_i32:
        tcg_out_qemu_st(s, args);
        break;
    case INDEX_op_qemu_st_i64:
        tcg_out_qemu_st(s, args);
        break;

    case INDEX_op_mov_i32:  /* Always emitted via tcg_out_mov.  */
    case INDEX_op_mov_i64:
    case INDEX_op_call:     /* Always emitted via tcg_out_call.  */
    case INDEX_op_exit_tb:  /* Always emitted via tcg_out_exit_tb.  */
    case INDEX_op_goto_tb:  /* Always emitted via tcg_out_goto_tb.  */
    default:
        g_assert_not_reached();
    }
}

static TCGConstraintSetIndex tcg_target_op_def(TCGOpcode op)
{
    switch (op) {
    case INDEX_op_goto_ptr:
        return C_O0_I1(r);

    case INDEX_op_st8_i32:
    case INDEX_op_st8_i64:
    case INDEX_op_st16_i32:
    case INDEX_op_st16_i64:
    case INDEX_op_st32_i64:
    case INDEX_op_st_i32:
    case INDEX_op_st_i64:
        return C_O0_I2(rZ, r);

    case INDEX_op_brcond_i32:
    case INDEX_op_brcond_i64:
        return C_O0_I2(rZ, rZ);

    case INDEX_op_qemu_st_i32:
    case INDEX_op_qemu_st_i64:
        return C_O0_I2(LZ, L);

    case INDEX_op_ext8s_i32:
    case INDEX_op_ext8s_i64:
    case INDEX_op_ext8u_i32:
    case INDEX_op_ext8u_i64:
    case INDEX_op_ext16s_i32:
    case INDEX_op_ext16s_i64:
    case INDEX_op_ext16u_i32:
    case INDEX_op_ext16u_i64:
    case INDEX_op_ext32s_i64:
    case INDEX_op_ext32u_i64:
    case INDEX_op_extu_i32_i64:
    case INDEX_op_extrl_i64_i32:
    case INDEX_op_extrh_i64_i32:
    case INDEX_op_ext_i32_i64:
    case INDEX_op_not_i32:
    case INDEX_op_not_i64:
    case INDEX_op_extract_i32:
    case INDEX_op_extract_i64:
    case INDEX_op_bswap16_i32:
    case INDEX_op_bswap16_i64:
    case INDEX_op_bswap32_i32:
    case INDEX_op_bswap32_i64:
    case INDEX_op_bswap64_i64:
    case INDEX_op_ld8s_i32:
    case INDEX_op_ld8s_i64:
    case INDEX_op_ld8u_i32:
    case INDEX_op_ld8u_i64:
    case INDEX_op_ld16s_i32:
    case INDEX_op_ld16s_i64:
    case INDEX_op_ld16u_i32:
    case INDEX_op_ld16u_i64:
    case INDEX_op_ld32s_i64:
    case INDEX_op_ld32u_i64:
    case INDEX_op_ld_i32:
    case INDEX_op_ld_i64:
        return C_O1_I1(r, r);

    case INDEX_op_qemu_ld_i32:
    case INDEX_op_qemu_ld_i64:
        return C_O1_I1(r, L);

    case INDEX_op_andc_i32:
    case INDEX_op_andc_i64:
    case INDEX_op_orc_i32:
    case INDEX_op_orc_i64:
        /*
         * LoongArch insns for these ops don't have reg-imm forms, but we
         * can express using andi/ori if ~constant satisfies
         * TCG_CT_CONST_U12.
         */
        return C_O1_I2(r, r, rC);

    case INDEX_op_shl_i32:
    case INDEX_op_shl_i64:
    case INDEX_op_shr_i32:
    case INDEX_op_shr_i64:
    case INDEX_op_sar_i32:
    case INDEX_op_sar_i64:
    case INDEX_op_rotl_i32:
    case INDEX_op_rotl_i64:
    case INDEX_op_rotr_i32:
    case INDEX_op_rotr_i64:
        return C_O1_I2(r, r, ri);

    case INDEX_op_add_i32:
        return C_O1_I2(r, r, ri);
    case INDEX_op_add_i64:
        return C_O1_I2(r, r, rJ);

    case INDEX_op_and_i32:
    case INDEX_op_and_i64:
    case INDEX_op_nor_i32:
    case INDEX_op_nor_i64:
    case INDEX_op_or_i32:
    case INDEX_op_or_i64:
    case INDEX_op_xor_i32:
    case INDEX_op_xor_i64:
        /* LoongArch reg-imm bitops have their imms ZERO-extended */
        return C_O1_I2(r, r, rU);

    case INDEX_op_clz_i32:
    case INDEX_op_clz_i64:
    case INDEX_op_ctz_i32:
    case INDEX_op_ctz_i64:
        return C_O1_I2(r, r, rW);

    case INDEX_op_setcond_i32:
    case INDEX_op_setcond_i64:
        return C_O1_I2(r, r, rZ);

    case INDEX_op_deposit_i32:
    case INDEX_op_deposit_i64:
        /* Must deposit into the same register as input */
        return C_O1_I2(r, 0, rZ);

    case INDEX_op_sub_i32:
        return C_O1_I2(r, rZ, ri);
    case INDEX_op_sub_i64:
        return C_O1_I2(r, rZ, rJ);

    case INDEX_op_mul_i32:
    case INDEX_op_mul_i64:
    case INDEX_op_mulsh_i32:
    case INDEX_op_mulsh_i64:
    case INDEX_op_muluh_i32:
    case INDEX_op_muluh_i64:
    case INDEX_op_div_i32:
    case INDEX_op_div_i64:
    case INDEX_op_divu_i32:
    case INDEX_op_divu_i64:
    case INDEX_op_rem_i32:
    case INDEX_op_rem_i64:
    case INDEX_op_remu_i32:
    case INDEX_op_remu_i64:
        return C_O1_I2(r, rZ, rZ);

    default:
        g_assert_not_reached();
    }
}

static const int tcg_target_callee_save_regs[] = {
    TCG_REG_S0,     /* used for the global env (TCG_AREG0) */
    TCG_REG_S1,
    TCG_REG_S2,
    TCG_REG_S3,
    TCG_REG_S4,
    TCG_REG_S5,
    TCG_REG_S6,
    TCG_REG_S7,
    TCG_REG_S8,
    TCG_REG_S9,
    TCG_REG_RA,     /* should be last for ABI compliance */
};

/* Stack frame parameters.  */
#define REG_SIZE   (TCG_TARGET_REG_BITS / 8)
#define SAVE_SIZE  ((int)ARRAY_SIZE(tcg_target_callee_save_regs) * REG_SIZE)
#define TEMP_SIZE  (CPU_TEMP_BUF_NLONGS * (int)sizeof(long))
#define FRAME_SIZE ((TCG_STATIC_CALL_ARGS_SIZE + TEMP_SIZE + SAVE_SIZE \
                     + TCG_TARGET_STACK_ALIGN - 1) \
                    & -TCG_TARGET_STACK_ALIGN)
#define SAVE_OFS   (TCG_STATIC_CALL_ARGS_SIZE + TEMP_SIZE)

/* We're expecting to be able to use an immediate for frame allocation.  */
QEMU_BUILD_BUG_ON(FRAME_SIZE > 0x7ff);

/* Generate global QEMU prologue and epilogue code */
static void tcg_target_qemu_prologue(TCGContext *s)
{
    int i;

    tcg_set_frame(s, TCG_REG_SP, TCG_STATIC_CALL_ARGS_SIZE, TEMP_SIZE);

    /* TB prologue */
    tcg_out_opc_addi_d(s, TCG_REG_SP, TCG_REG_SP, -FRAME_SIZE);
    for (i = 0; i < ARRAY_SIZE(tcg_target_callee_save_regs); i++) {
        tcg_out_st(s, TCG_TYPE_REG, tcg_target_callee_save_regs[i],
                   TCG_REG_SP, SAVE_OFS + i * REG_SIZE);
    }

#if !defined(CONFIG_SOFTMMU)
    if (USE_GUEST_BASE) {
        tcg_out_movi(s, TCG_TYPE_PTR, TCG_GUEST_BASE_REG, guest_base);
        tcg_regset_set_reg(s->reserved_regs, TCG_GUEST_BASE_REG);
    }
#endif

    /* Call generated code */
    tcg_out_mov(s, TCG_TYPE_PTR, TCG_AREG0, tcg_target_call_iarg_regs[0]);
    tcg_out_opc_jirl(s, TCG_REG_ZERO, tcg_target_call_iarg_regs[1], 0);

    /* Return path for goto_ptr. Set return value to 0 */
    tcg_code_gen_epilogue = tcg_splitwx_to_rx(s->code_ptr);
    tcg_out_mov(s, TCG_TYPE_REG, TCG_REG_A0, TCG_REG_ZERO);

    /* TB epilogue */
    tb_ret_addr = tcg_splitwx_to_rx(s->code_ptr);
    for (i = 0; i < ARRAY_SIZE(tcg_target_callee_save_regs); i++) {
        tcg_out_ld(s, TCG_TYPE_REG, tcg_target_callee_save_regs[i],
                   TCG_REG_SP, SAVE_OFS + i * REG_SIZE);
    }

    tcg_out_opc_addi_d(s, TCG_REG_SP, TCG_REG_SP, FRAME_SIZE);
    tcg_out_opc_jirl(s, TCG_REG_ZERO, TCG_REG_RA, 0);
}

static void tcg_target_init(TCGContext *s)
{
    tcg_target_available_regs[TCG_TYPE_I32] = ALL_GENERAL_REGS;
    tcg_target_available_regs[TCG_TYPE_I64] = ALL_GENERAL_REGS;

    tcg_target_call_clobber_regs = ALL_GENERAL_REGS;
    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_S0);
    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_S1);
    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_S2);
    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_S3);
    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_S4);
    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_S5);
    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_S6);
    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_S7);
    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_S8);
    tcg_regset_reset_reg(tcg_target_call_clobber_regs, TCG_REG_S9);

    s->reserved_regs = 0;
    tcg_regset_set_reg(s->reserved_regs, TCG_REG_ZERO);
    tcg_regset_set_reg(s->reserved_regs, TCG_REG_TMP0);
    tcg_regset_set_reg(s->reserved_regs, TCG_REG_TMP1);
    tcg_regset_set_reg(s->reserved_regs, TCG_REG_TMP2);
    tcg_regset_set_reg(s->reserved_regs, TCG_REG_SP);
    tcg_regset_set_reg(s->reserved_regs, TCG_REG_TP);
    tcg_regset_set_reg(s->reserved_regs, TCG_REG_RESERVED);
}

typedef struct {
    DebugFrameHeader h;
    uint8_t fde_def_cfa[4];
    uint8_t fde_reg_ofs[ARRAY_SIZE(tcg_target_callee_save_regs) * 2];
} DebugFrame;

#define ELF_HOST_MACHINE EM_LOONGARCH

static const DebugFrame debug_frame = {
    .h.cie.len = sizeof(DebugFrameCIE) - 4, /* length after .len member */
    .h.cie.id = -1,
    .h.cie.version = 1,
    .h.cie.code_align = 1,
    .h.cie.data_align = -(TCG_TARGET_REG_BITS / 8) & 0x7f, /* sleb128 */
    .h.cie.return_column = TCG_REG_RA,

    /* Total FDE size does not include the "len" member.  */
    .h.fde.len = sizeof(DebugFrame) - offsetof(DebugFrame, h.fde.cie_offset),

    .fde_def_cfa = {
        12, TCG_REG_SP,                 /* DW_CFA_def_cfa sp, ...  */
        (FRAME_SIZE & 0x7f) | 0x80,     /* ... uleb128 FRAME_SIZE */
        (FRAME_SIZE >> 7)
    },
    .fde_reg_ofs = {
        0x80 + 23, 11,                  /* DW_CFA_offset, s0, -88 */
        0x80 + 24, 10,                  /* DW_CFA_offset, s1, -80 */
        0x80 + 25, 9,                   /* DW_CFA_offset, s2, -72 */
        0x80 + 26, 8,                   /* DW_CFA_offset, s3, -64 */
        0x80 + 27, 7,                   /* DW_CFA_offset, s4, -56 */
        0x80 + 28, 6,                   /* DW_CFA_offset, s5, -48 */
        0x80 + 29, 5,                   /* DW_CFA_offset, s6, -40 */
        0x80 + 30, 4,                   /* DW_CFA_offset, s7, -32 */
        0x80 + 31, 3,                   /* DW_CFA_offset, s8, -24 */
        0x80 + 22, 2,                   /* DW_CFA_offset, s9, -16 */
        0x80 + 1 , 1,                   /* DW_CFA_offset, ra, -8 */
    }
};

void tcg_register_jit(const void *buf, size_t buf_size)
{
    tcg_register_jit_int(buf, buf_size, &debug_frame, sizeof(debug_frame));
}