fxos/lib/lang.cpp

//---------------------------------------------------------------------------//
//  1100101 |_ mov #0, r4         __                                         //
//     11   |_ <0xb380 %5c4>     / _|_ _____ ___                             //
//     0110 |_ 3.50 -> 3.60     |  _\ \ / _ (_-<                             //
//          |_ base# + offset   |_| /_\_\___/__/                             //
//---------------------------------------------------------------------------//

#include <fxos/lang.h>
#include <fxos/util/format.h>
#include <string>
#include <cstring>
#include <cassert>

namespace FxOS {

//---
// CPU registers
//---

// clang-format off
char const *regnames[] = {
    "r0", "r1", "r2", "r3", "r4", "r5", "r6", "r7",
    "r0_bank", "r1_bank", "r2_bank", "r3_bank",
    "r4_bank", "r5_bank", "r6_bank", "r7_bank",
    "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15",
    "mach", "macl", "pr", "pc",
    "sr", "ssr", "spc", "gbr", "vbr", "dbr", "sgr"
};
// clang-format on

/* Construction from string */
CpuRegister::CpuRegister(std::string name)
{
    int regcount = (sizeof regnames / sizeof regnames[0]);
    char const *name_c = name.c_str();

    for(int i = 0; i < regcount; i++) {
        if(!strcmp(regnames[i], name_c)) {
            m_name = CpuRegisterName(i + 1);
            return;
        }
    }

    m_name = CpuRegister::UNDEFINED;
}

/* Conversion to string */
std::string CpuRegister::str() const noexcept
{
    int regcount = (sizeof regnames / sizeof regnames[0]);
    int i = m_name - 1;

    if(i < 0 || i >= regcount)
        return format("<Register%d>", i + 1);
    return regnames[i];
}

//---
// Instruction operands
//---

AsmOperand::AsmOperand(Kind kind, CpuRegister base):
    m_kind {kind}, m_base {base}
{
}

AsmOperand::AsmOperand(int disp, i8 opsize, CpuRegister base):
    m_kind {StructDeref}, m_base {base}, m_opsize {opsize}, m_disp_imm {disp}
{
}

AsmOperand::AsmOperand(CpuRegister index, CpuRegister base):
    m_kind {ArrayDeref}, m_base {base}, m_index {index}
{
}

AsmOperand::AsmOperand(Kind kind, int disp_imm, i8 opsize):
    m_kind {kind}, m_opsize {opsize}, m_disp_imm {disp_imm}
{
}

std::string AsmOperand::str() const
{
    switch(m_kind) {
    case Reg:
        return m_base.str();
    case Deref:
        return format("@%s", m_base.str());
    case PostInc:
        return format("@%s+", m_base.str());
    case PreDec:
        return format("@-%s", m_base.str());
    case StructDeref:
        return format("@(%d,%s)", m_disp_imm, m_base.str().c_str());
    case ArrayDeref:
        return format("@(%s,%s)", m_index.str().c_str(), m_base.str().c_str());
    case PcRel:
        return format("@(%d,pc)", m_disp_imm);
    case PcJump:
        return format("pc+%d", m_disp_imm);
    case PcAddr:
        return format("pc+%u", m_disp_imm);
    case Imm:
        return format("#%d", m_disp_imm);
    default:
        return "(invalid)";
    }
}

u32 AsmOperand::getPCRelativeTarget(u32 pc, int size) const
{
    size = size + (size == 0);

    if(m_kind == AsmOperand::PcRel)
        return (pc & -size) + 4 + m_disp_imm;
    if(m_kind == AsmOperand::PcJump)
        return pc + 4 + m_disp_imm;
    if(m_kind == AsmOperand::PcAddr)
        return (pc & -4) + 4 + m_disp_imm;

    /* SH3 manual says that mova uses the target address of the jump when
       in a delay slot. SH4AL-DSP makes it invalid. Supporting this would
       be very tricky since the target PC is often dynamic (eg. rts). */

    return -1;
}

//---
//  Instruction management
//---

static char const *instructionMnemonics[] = {
#define GENDEFS_INSN(NAME, STR) [AsmInstruction::SH_##NAME] = STR,
#include "gendefs/insn.h"
#undef GENDEFS_INSN
};

AsmInstruction::AsmInstruction(u32 encoding, char const *mnemonic, int tags,
    int opCount, AsmOperand op1, AsmOperand op2):
    m_encoding {encoding},
    m_opsize {0}, m_opCount {(u8)opCount}, m_tags {(u16)tags}, m_ops {op1, op2}
{
    std::string mn {mnemonic};

    if(mn.ends_with(".b"))
        m_opsize = 1;
    else if(mn.ends_with(".w"))
        m_opsize = 2;
    else if(mn.ends_with(".l"))
        m_opsize = 4;

    if(m_opsize != 0)
        mn = mn.substr(0, mn.size() - 2);

    int i;
    for(i = 0; i < SH_MAX; i++) {
        if(mn == instructionMnemonics[i]) {
            m_opcode = i;
            break;
        }
    }
    assert(i < SH_MAX && "AsmInstruction with unknown opcode string");
}

char const *AsmInstruction::mnemonic() const
{
    assert(m_opcode < SH_MAX);
    return instructionMnemonics[m_opcode];
}

u32 AsmInstruction::getPCRelativeTarget(u32 pc) const
{
    for(AsmOperand const &op: operands()) {
        if(op.usesPCRelativeAddressing())
            return op.getPCRelativeTarget(pc, m_opsize);
    }
    return -1;
}

} /* namespace FxOS */