/*
 * Copyright (c) 2023, Andreas Kling <kling@serenityos.org>
 *
 * SPDX-License-Identifier: BSD-2-Clause
 */

#pragma once

#include <AK/Platform.h>
#include <AK/Vector.h>

#if ARCH(X86_64)

namespace JIT {

struct Assembler {
    Assembler(Vector<u8>& output)
        : m_output(output)
    {
    }

    Vector<u8>& m_output;

    enum class Reg {
        RAX = 0,
        RCX = 1,
        RDX = 2,
        RBX = 3,
        RSP = 4,
        RBP = 5,
        RSI = 6,
        RDI = 7,
        R8 = 8,
        R9 = 9,
        R10 = 10,
        R11 = 11,
        R12 = 12,
        R13 = 13,
        R14 = 14,
        R15 = 15,
    };

    struct Operand {
        enum class Type {
            Reg,
            Imm,
            Mem64BaseAndOffset,
        };

        Type type {};

        Reg reg {};
        u64 offset_or_immediate { 0 };

        static Operand Register(Reg reg)
        {
            Operand operand;
            operand.type = Type::Reg;
            operand.reg = reg;
            return operand;
        }

        static Operand Imm(u64 imm)
        {
            Operand operand;
            operand.type = Type::Imm;
            operand.offset_or_immediate = imm;
            return operand;
        }

        static Operand Mem64BaseAndOffset(Reg base, u64 offset)
        {
            Operand operand;
            operand.type = Type::Mem64BaseAndOffset;
            operand.reg = base;
            operand.offset_or_immediate = offset;
            return operand;
        }

        bool fits_in_u8() const
        {
            VERIFY(type == Type::Imm);
            return offset_or_immediate <= NumericLimits<u8>::max();
        }
        bool fits_in_u32() const
        {
            VERIFY(type == Type::Imm);
            return offset_or_immediate <= NumericLimits<u32>::max();
        }
        bool fits_in_i8() const
        {
            VERIFY(type == Type::Imm);
            return (offset_or_immediate <= NumericLimits<i8>::max()) || (((~offset_or_immediate) & NumericLimits<i8>::min()) == 0);
        }
        bool fits_in_i32() const
        {
            VERIFY(type == Type::Imm);
            return (offset_or_immediate <= NumericLimits<i32>::max()) || (((~offset_or_immediate) & NumericLimits<i32>::min()) == 0);
        }
    };

    enum class Condition {
        EqualTo = 0x4,
        NotEqualTo = 0x5,
        UnsignedGreaterThan = 0x7,
        UnsignedGreaterThanOrEqualTo = 0x3,
        UnsignedLessThan = 0x2,
        UnsignedLessThanOrEqualTo = 0x6,
        SignedGreaterThan = 0xF,
        SignedGreaterThanOrEqualTo = 0xD,
        SignedLessThan = 0xC,
        SignedLessThanOrEqualTo = 0xE,
    };

    static constexpr u8 encode_reg(Reg reg)
    {
        return to_underlying(reg) & 0x7;
    }

    void shift_right(Operand dst, Operand count)
    {
        VERIFY(dst.type == Operand::Type::Reg);
        VERIFY(count.type == Operand::Type::Imm);
        VERIFY(count.fits_in_u8());
        emit8(0x48 | ((to_underlying(dst.reg) >= 8) ? 1 << 0 : 0));
        emit8(0xc1);
        emit8(0xe8 | encode_reg(dst.reg));
        emit8(count.offset_or_immediate);
    }

    enum class Patchable {
        Yes,
        No,
    };

    void mov(Operand dst, Operand src, Patchable patchable = Patchable::No)
    {
        if (dst.type == Operand::Type::Reg && src.type == Operand::Type::Reg) {
            if (src.reg == dst.reg)
                return;
            emit8(0x48
                | ((to_underlying(src.reg) >= 8) ? 1 << 2 : 0)
                | ((to_underlying(dst.reg) >= 8) ? 1 << 0 : 0));
            emit8(0x89);
            emit8(0xc0 | (encode_reg(src.reg) << 3) | encode_reg(dst.reg));
            return;
        }

        if (dst.type == Operand::Type::Reg && src.type == Operand::Type::Imm) {
            if (patchable == Patchable::No) {
                if (src.offset_or_immediate == 0) {
                    // xor dst, dst
                    emit8(0x48 | ((to_underlying(dst.reg) >= 8) ? (1 << 0 | 1 << 2) : 0));
                    emit8(0x31);
                    emit8(0xc0 | (encode_reg(dst.reg) << 3) | encode_reg(dst.reg));
                    return;
                }
                if (src.fits_in_u32()) {
                    if (dst.reg > Reg::RDI)
                        emit8(0x41);
                    emit8(0xb8 | encode_reg(dst.reg));
                    emit32(src.offset_or_immediate);
                    return;
                }
            }
            emit8(0x48 | ((to_underlying(dst.reg) >= 8) ? 1 << 0 : 0));
            emit8(0xb8 | encode_reg(dst.reg));
            emit64(src.offset_or_immediate);
            return;
        }

        if (dst.type == Operand::Type::Mem64BaseAndOffset && src.type == Operand::Type::Reg) {
            emit8(0x48
                | ((to_underlying(src.reg) >= 8) ? 1 << 2 : 0)
                | ((to_underlying(dst.reg) >= 8) ? 1 << 0 : 0));
            emit8(0x89);
            if (dst.reg <= Reg::RDI && dst.offset_or_immediate == 0) {
                emit8(0x00 | (encode_reg(src.reg) << 3) | encode_reg(dst.reg));
            } else if (dst.offset_or_immediate <= 127) {
                emit8(0x40 | (encode_reg(src.reg) << 3) | encode_reg(dst.reg));
                emit8(dst.offset_or_immediate);
            } else {
                emit8(0x80 | (encode_reg(src.reg) << 3) | encode_reg(dst.reg));
                emit32(dst.offset_or_immediate);
            }
            return;
        }

        if (dst.type == Operand::Type::Reg && src.type == Operand::Type::Mem64BaseAndOffset) {
            emit8(0x48
                | ((to_underlying(dst.reg) >= 8) ? 1 << 2 : 0)
                | ((to_underlying(src.reg) >= 8) ? 1 << 0 : 0));
            emit8(0x8b);
            if (src.reg <= Reg::RDI && src.offset_or_immediate == 0) {
                emit8(0x00 | (encode_reg(dst.reg) << 3) | encode_reg(src.reg));
            } else if (src.offset_or_immediate <= 127) {
                emit8(0x40 | (encode_reg(dst.reg) << 3) | encode_reg(src.reg));
                emit8(src.offset_or_immediate);
            } else {
                emit8(0x80 | (encode_reg(dst.reg) << 3) | encode_reg(src.reg));
                emit32(src.offset_or_immediate);
            }
            return;
        }

        VERIFY_NOT_REACHED();
    }

    void emit8(u8 value)
    {
        m_output.append(value);
    }

    void emit32(u32 value)
    {
        m_output.append((value >> 0) & 0xff);
        m_output.append((value >> 8) & 0xff);
        m_output.append((value >> 16) & 0xff);
        m_output.append((value >> 24) & 0xff);
    }

    void emit64(u64 value)
    {
        m_output.append((value >> 0) & 0xff);
        m_output.append((value >> 8) & 0xff);
        m_output.append((value >> 16) & 0xff);
        m_output.append((value >> 24) & 0xff);
        m_output.append((value >> 32) & 0xff);
        m_output.append((value >> 40) & 0xff);
        m_output.append((value >> 48) & 0xff);
        m_output.append((value >> 56) & 0xff);
    }

    struct Label {
        Optional<size_t> offset_of_label_in_instruction_stream;
        Vector<size_t> jump_slot_offsets_in_instruction_stream;

        void add_jump(Assembler& assembler, size_t offset)
        {
            jump_slot_offsets_in_instruction_stream.append(offset);
            if (offset_of_label_in_instruction_stream.has_value())
                link_jump(assembler, offset);
        }

        void link(Assembler& assembler)
        {
            link_to(assembler, assembler.m_output.size());
        }

        void link_to(Assembler& assembler, size_t link_offset)
        {
            VERIFY(!offset_of_label_in_instruction_stream.has_value());
            offset_of_label_in_instruction_stream = link_offset;
            for (auto offset_in_instruction_stream : jump_slot_offsets_in_instruction_stream)
                link_jump(assembler, offset_in_instruction_stream);
        }

    private:
        void link_jump(Assembler& assembler, size_t offset_in_instruction_stream)
        {
            auto offset = offset_of_label_in_instruction_stream.value() - offset_in_instruction_stream;
            auto jump_slot = offset_in_instruction_stream - 4;
            assembler.m_output[jump_slot + 0] = (offset >> 0) & 0xff;
            assembler.m_output[jump_slot + 1] = (offset >> 8) & 0xff;
            assembler.m_output[jump_slot + 2] = (offset >> 16) & 0xff;
            assembler.m_output[jump_slot + 3] = (offset >> 24) & 0xff;
        }
    };

    [[nodiscard]] Label jump()
    {
        // jmp target (RIP-relative 32-bit offset)
        emit8(0xe9);
        emit32(0xdeadbeef);
        Assembler::Label label {};
        label.add_jump(*this, m_output.size());
        return label;
    }

    void jump(Label& label)
    {
        // jmp target (RIP-relative 32-bit offset)
        emit8(0xe9);
        emit32(0xdeadbeef);
        label.add_jump(*this, m_output.size());
    }

    void jump(Operand op)
    {
        if (op.type == Operand::Type::Reg) {
            if (to_underlying(op.reg) >= 8)
                emit8(0x41);
            emit8(0xff);
            emit8(0xe0 | encode_reg(op.reg));
        } else {
            VERIFY_NOT_REACHED();
        }
    }

    void verify_not_reached()
    {
        // ud2
        emit8(0x0f);
        emit8(0x0b);
    }

    void cmp(Operand lhs, Operand rhs)
    {
        if (rhs.type == Operand::Type::Imm && rhs.offset_or_immediate == 0) {
            test(lhs, lhs);
        } else if (lhs.type == Operand::Type::Reg && rhs.type == Operand::Type::Reg) {
            emit8(0x48
                | ((to_underlying(rhs.reg) >= 8) ? 1 << 2 : 0)
                | ((to_underlying(lhs.reg) >= 8) ? 1 << 0 : 0));
            emit8(0x39);
            emit8(0xc0 | (encode_reg(rhs.reg) << 3) | encode_reg(lhs.reg));
        } else if (lhs.type == Operand::Type::Reg && rhs.type == Operand::Type::Imm && rhs.fits_in_i8()) {
            emit8(0x48 | ((to_underlying(lhs.reg) >= 8) ? 1 << 0 : 0));
            emit8(0x83);
            emit8(0xf8 | encode_reg(lhs.reg));
            emit8(rhs.offset_or_immediate);
        } else if (lhs.type == Operand::Type::Reg && rhs.type == Operand::Type::Imm && rhs.fits_in_i32()) {
            emit8(0x48 | ((to_underlying(lhs.reg) >= 8) ? 1 << 0 : 0));
            emit8(0x81);
            emit8(0xf8 | encode_reg(lhs.reg));
            emit32(rhs.offset_or_immediate);
        } else {
            VERIFY_NOT_REACHED();
        }
    }

    void test(Operand lhs, Operand rhs)
    {
        if (lhs.type == Operand::Type::Reg && rhs.type == Operand::Type::Reg) {
            emit8(0x48
                | ((to_underlying(rhs.reg) >= 8) ? 1 << 2 : 0)
                | ((to_underlying(lhs.reg) >= 8) ? 1 << 0 : 0));
            emit8(0x85);
            emit8(0xc0 | (encode_reg(rhs.reg) << 3) | encode_reg(lhs.reg));
        } else if (lhs.type == Operand::Type::Reg && rhs.type == Operand::Type::Imm) {
            VERIFY(rhs.fits_in_i32());
            emit8(0x48 | ((to_underlying(lhs.reg) >= 8) ? 1 << 0 : 0));
            emit8(0xf7);
            emit8(0xc0 | encode_reg(lhs.reg));
            emit32(rhs.offset_or_immediate);
        } else {
            VERIFY_NOT_REACHED();
        }
    }

    void jump_if(Operand lhs, Condition condition, Operand rhs, Label& label)
    {
        cmp(lhs, rhs);

        emit8(0x0F);
        emit8(0x80 | to_underlying(condition));
        emit32(0xdeadbeef);
        label.add_jump(*this, m_output.size());
    }

    void sign_extend_32_to_64_bits(Reg reg)
    {
        // movsxd (reg as 64-bit), (reg as 32-bit)
        emit8(0x48 | ((to_underlying(reg) >= 8) ? 1 << 0 : 0));
        emit8(0x63);
        emit8(0xc0 | (encode_reg(reg) << 3) | encode_reg(reg));
    }

    void bitwise_and(Operand dst, Operand src)
    {
        // and dst,src
        if (dst.type == Operand::Type::Reg && src.type == Operand::Type::Reg) {
            emit8(0x48
                | ((to_underlying(src.reg) >= 8) ? 1 << 2 : 0)
                | ((to_underlying(dst.reg) >= 8) ? 1 << 0 : 0));
            emit8(0x21);
            emit8(0xc0 | (encode_reg(src.reg) << 3) | encode_reg(dst.reg));
        } else if (dst.type == Operand::Type::Reg && src.type == Operand::Type::Imm && src.fits_in_i8()) {
            emit8(0x48 | ((to_underlying(dst.reg) >= 8) ? 1 << 0 : 0));
            emit8(0x83);
            emit8(0xe0 | encode_reg(dst.reg));
            emit8(src.offset_or_immediate);
        } else if (dst.type == Operand::Type::Reg && src.type == Operand::Type::Imm && src.fits_in_i32()) {
            emit8(0x48 | ((to_underlying(dst.reg) >= 8) ? 1 << 0 : 0));
            emit8(0x81);
            emit8(0xe0 | encode_reg(dst.reg));
            emit32(src.offset_or_immediate);
        } else {
            VERIFY_NOT_REACHED();
        }
    }

    void bitwise_or(Operand dst, Operand src)
    {
        // or dst,src
        if (dst.type == Operand::Type::Reg && src.type == Operand::Type::Reg) {
            emit8(0x48
                | ((to_underlying(src.reg) >= 8) ? 1 << 2 : 0)
                | ((to_underlying(dst.reg) >= 8) ? 1 << 0 : 0));
            emit8(0x09);
            emit8(0xc0 | (encode_reg(src.reg) << 3) | encode_reg(dst.reg));
        } else if (dst.type == Operand::Type::Reg && src.type == Operand::Type::Imm && src.fits_in_i8()) {
            emit8(0x48 | ((to_underlying(dst.reg) >= 8) ? 1 << 0 : 0));
            emit8(0x83);
            emit8(0xc8 | encode_reg(dst.reg));
            emit8(src.offset_or_immediate);
        } else if (dst.type == Operand::Type::Reg && src.type == Operand::Type::Imm && src.fits_in_i32()) {
            emit8(0x48 | ((to_underlying(dst.reg) >= 8) ? 1 << 0 : 0));
            emit8(0x81);
            emit8(0xc8 | encode_reg(dst.reg));
            emit32(src.offset_or_immediate);
        } else {
            VERIFY_NOT_REACHED();
        }
    }

    void enter()
    {
        push_callee_saved_registers();

        push(Operand::Register(Reg::RBP));
        mov(Operand::Register(Reg::RBP), Operand::Register(Reg::RSP));
    }

    void exit()
    {
        // leave
        emit8(0xc9);

        pop_callee_saved_registers();

        // ret
        emit8(0xc3);
    }

    void push_callee_saved_registers()
    {
        // FIXME: Don't push RBX twice :^)
        push(Operand::Register(Reg::RBX));
        push(Operand::Register(Reg::RBX));
        push(Operand::Register(Reg::R12));
        push(Operand::Register(Reg::R13));
        push(Operand::Register(Reg::R14));
        push(Operand::Register(Reg::R15));
    }

    void pop_callee_saved_registers()
    {
        pop(Operand::Register(Reg::R15));
        pop(Operand::Register(Reg::R14));
        pop(Operand::Register(Reg::R13));
        pop(Operand::Register(Reg::R12));

        // FIXME: Don't pop RBX twice :^)
        pop(Operand::Register(Reg::RBX));
        pop(Operand::Register(Reg::RBX));
    }

    void push(Operand op)
    {
        if (op.type == Operand::Type::Reg) {
            if (to_underlying(op.reg) >= 8)
                emit8(0x49);
            emit8(0x50 | encode_reg(op.reg));
        } else if (op.type == Operand::Type::Imm) {
            if (op.fits_in_i8()) {
                emit8(0x6a);
                emit8(op.offset_or_immediate);
            } else if (op.fits_in_i32()) {
                emit8(0x68);
                emit32(op.offset_or_immediate);
            } else {
                VERIFY_NOT_REACHED();
            }
        } else {
            VERIFY_NOT_REACHED();
        }
    }

    void pop(Operand op)
    {
        if (op.type == Operand::Type::Reg) {
            if (to_underlying(op.reg) >= 8)
                emit8(0x49);
            emit8(0x58 | encode_reg(op.reg));
        } else {
            VERIFY_NOT_REACHED();
        }
    }

    void add(Operand dst, Operand src)
    {
        if (dst.type == Operand::Type::Reg && src.type == Operand::Type::Reg) {
            emit8(0x48
                | ((to_underlying(src.reg) >= 8) ? 1 << 2 : 0)
                | ((to_underlying(dst.reg) >= 8) ? 1 << 0 : 0));
            emit8(0x01);
            emit8(0xc0 | (encode_reg(src.reg) << 3) | encode_reg(dst.reg));
        } else if (dst.type == Operand::Type::Reg && src.type == Operand::Type::Imm && src.fits_in_i8()) {
            emit8(0x48 | ((to_underlying(dst.reg) >= 8) ? 1 << 0 : 0));
            emit8(0x83);
            emit8(0xc0 | encode_reg(dst.reg));
            emit8(src.offset_or_immediate);
        } else if (dst.type == Operand::Type::Reg && src.type == Operand::Type::Imm && src.fits_in_i32()) {
            emit8(0x48 | ((to_underlying(dst.reg) >= 8) ? 1 << 0 : 0));
            emit8(0x81);
            emit8(0xc0 | encode_reg(dst.reg));
            emit32(src.offset_or_immediate);
        } else {
            VERIFY_NOT_REACHED();
        }
    }

    void add32(Operand dst, Operand src, Optional<Label&> label)
    {
        if (dst.type == Operand::Type::Reg && to_underlying(dst.reg) < 8 && src.type == Operand::Type::Reg && to_underlying(src.reg) < 8) {
            emit8(0x01);
            emit8(0xc0 | (encode_reg(src.reg) << 3) | encode_reg(dst.reg));
        } else if (dst.type == Operand::Type::Reg && to_underlying(dst.reg) < 8 && src.type == Operand::Type::Imm && src.fits_in_i8()) {
            emit8(0x83);
            emit8(0xc0 | encode_reg(dst.reg));
            emit8(src.offset_or_immediate);
        } else if (dst.type == Operand::Type::Reg && to_underlying(dst.reg) < 8 && src.type == Operand::Type::Imm && src.fits_in_i32()) {
            emit8(0x81);
            emit8(0xc0 | encode_reg(dst.reg));
            emit32(src.offset_or_immediate);
        } else {
            VERIFY_NOT_REACHED();
        }

        if (label.has_value()) {
            // jo label (RIP-relative 32-bit offset)
            emit8(0x0f);
            emit8(0x80);
            emit32(0xdeadbeef);
            label->add_jump(*this, m_output.size());
        }
    }

    void sub(Operand dst, Operand src)
    {
        if (dst.type == Operand::Type::Reg && src.type == Operand::Type::Reg) {
            emit8(0x48
                | ((to_underlying(src.reg) >= 8) ? 1 << 2 : 0)
                | ((to_underlying(dst.reg) >= 8) ? 1 << 0 : 0));
            emit8(0x29);
            emit8(0xc0 | (encode_reg(src.reg) << 3) | encode_reg(dst.reg));
        } else if (dst.type == Operand::Type::Reg && src.type == Operand::Type::Imm && src.fits_in_i8()) {
            emit8(0x48 | ((to_underlying(dst.reg) >= 8) ? 1 << 0 : 0));
            emit8(0x83);
            emit8(0xe8 | encode_reg(dst.reg));
            emit8(src.offset_or_immediate);
        } else if (dst.type == Operand::Type::Reg && src.type == Operand::Type::Imm && src.fits_in_i32()) {
            emit8(0x48 | ((to_underlying(dst.reg) >= 8) ? 1 << 0 : 0));
            emit8(0x81);
            emit8(0xe8 | encode_reg(dst.reg));
            emit32(src.offset_or_immediate);
        } else {
            VERIFY_NOT_REACHED();
        }
    }

    // NOTE: It's up to the caller of this function to preserve registers as needed.
    void native_call(void* callee, Vector<Operand> const& stack_arguments = {})
    {
        // Preserve 16-byte stack alignment for non-even amount of stack-passed arguments
        if ((stack_arguments.size() % 2) == 1)
            push(Operand::Imm(0));
        for (auto const& stack_argument : stack_arguments.in_reverse())
            push(stack_argument);

        // load callee into RAX
        mov(Operand::Register(Reg::RAX), Operand::Imm(bit_cast<u64>(callee)));

        // call RAX
        emit8(0xff);
        emit8(0xd0);

        if (!stack_arguments.is_empty())
            add(Operand::Register(Reg::RSP), Operand::Imm(align_up_to(stack_arguments.size(), 2) * sizeof(void*)));
    }

    void trap()
    {
        // int3
        emit8(0xcc);
    }
};

}

#endif