"""Intel i860 disassembler for the VelociRender firmware.

Opcode/field assignments are those VALIDATED against our own AS860 .S<->.O pairs
(see i860-encoding.md / derive860.py); unknowns decode as `.word`. This is also
the front-end for an eventual i860 *emulator* that runs VREND.MNG directly.

Run `python dis860.py --validate` to check the decoder against the matching
source/object pairs (per-instruction base-mnemonic agreement)."""
import sys, struct, os, glob

REG = [f"r{i}" for i in range(32)]
FREG = [f"f{i}" for i in range(32)]
CTRL = {0: "fir", 1: "psr", 2: "dirbase", 3: "db", 4: "fsr", 5: "epsr"}

def s16(v): return v - 0x10000 if v & 0x8000 else v
def s26(v): return v - 0x4000000 if v & 0x2000000 else v

# ---- FP escape (op 0x12) sub-opcode[6:0] -> base name (validated where marked) ----
FP_SUB = {
    0x00: "r2p1", 0x01: "r2pt", 0x02: "r2ap1", 0x04: "i2p1", 0x05: "i2pt",
    0x06: "i2ap1", 0x07: "i2apt", 0x08: "rat1p2", 0x09: "m12apm",
    0x0a: "ra1p2", 0x0b: "m12ttpa", 0x0c: "iat1p2", 0x0d: "m12tpm",
    0x0e: "ia1p2", 0x0f: "m12tpa", 0x1c: "mimt1s2",
    0x20: "fmul", 0x21: "fmlow", 0x22: "frcp", 0x23: "frsqr",
    0x24: "fmul3", 0x28: "pfmam", 0x2c: "pfmsm",
    0x30: "fadd", 0x31: "fsub", 0x32: "fix", 0x34: "fgt", 0x35: "feq",
    0x3a: "ftrunc", 0x40: "fxfr", 0x49: "fiadd", 0x4d: "fisub",
    0x50: "faddp", 0x51: "faddz", 0x57: "fzchk", 0x5a: "form",
    0x5b: "fzchkl", 0x5f: "fnop",
}
# precision suffix from result/source-precision bits (bit8=R result-prec, bit7=S src-prec)
def fp_suffix(w):
    s = 'd' if (w >> 7) & 1 else 's'
    r = 'd' if (w >> 8) & 1 else 's'
    return f".{s}{r}"

def dec_fp(w):
    sub = w & 0x7f
    p = (w >> 10) & 1                 # P: pipelined
    src2 = (w >> 21) & 0x1f; dest = (w >> 16) & 0x1f; src1 = (w >> 11) & 0x1f
    base = FP_SUB.get(sub)
    if base is None:
        return (".fp?", f"{sub:#04x} {FREG[src1]},{FREG[src2]},{FREG[dest]}")
    if base == "fxfr":               # FP->int reg move
        return ("fxfr", f"{FREG[src1]},{REG[dest]}")
    if base == "fnop":
        return ("fnop", "")
    name = ("pf" if p else "f") + base[1:] if base[0] == 'f' else (("pf" if p else "") + base)
    # graphics dual-ops (r2p1 etc.) keep their name; only f* get the pf/f prefix
    if base[0] != 'f':
        name = base
    return (name + fp_suffix(w), f"{FREG[src1]},{FREG[src2]},{FREG[dest]}")

# ---- primary decode ----
# load/store encoding (ground truth: AS860 .S<->.O pairs + DNC.O):
#  - integer loads (0x00/0x01 ld.b, 0x04/0x05 ld.s|ld.l): dest=bits20:16;
#    ODD = 16-bit immediate offset, EVEN = register-indexed EA=src2+src1.
#    op 4/5 size flag = instr bit0 (0=.s 16-bit, 1=.l 32-bit).
#  - integer stores (0x03 st.b, 0x07 st.s|st.l): source reg = src1 (bits15:11),
#    offset SPLIT high=bits20:16 low=bits10:0; bit0 size flag as loads.
#  - FP fld (0x08/0x09) / fst (0x0a/0x0b): FP reg = dest field for BOTH;
#    ODD = flat s16 imm, EVEN = indexed. Flags: bit0=auto-increment,
#    bit1: 1=.l, 0=.d; bit2 (with bit1=0) = .q.
MEM = {0x00, 0x01, 0x03, 0x04, 0x05, 0x07, 0x08, 0x09, 0x0a, 0x0b}
ARITH = {0x20: "addu", 0x22: "subu", 0x24: "adds", 0x26: "subs",
         0x28: "shl", 0x2a: "shr", 0x2c: "shrd", 0x2e: "shra",
         0x30: "and", 0x32: "andh", 0x34: "andnot", 0x36: "andnoth",
         0x38: "or", 0x3a: "orh", 0x3c: "xor", 0x3e: "xorh"}
CTRLB = {0x1a: "br", 0x1b: "call", 0x1c: "bc", 0x1d: "bc.t",
         0x1e: "bnc", 0x1f: "bnc.t"}

def decode(w, addr):
    op = (w >> 26) & 0x3f
    src2 = (w >> 21) & 0x1f; dest = (w >> 16) & 0x1f; src1 = (w >> 11) & 0x1f
    imm = w & 0xffff
    # loads/stores
    if op in MEM:
        if op in (0x00, 0x01, 0x04, 0x05):          # integer loads
            m = "ld.b" if op < 0x04 else ("ld.l" if (w & 1) else "ld.s")
            if op & 1:
                mask = 0xffff if op == 0x01 else (0xfffc if (w & 1) else 0xfffe)
                return m, f"{s16(imm & mask):#x}({REG[src2]}),{REG[dest]}"
            return m, f"{REG[src1]}({REG[src2]}),{REG[dest]}"
        if op in (0x03, 0x07):                       # integer stores (split offset)
            off = ((w >> 16) & 0x1f) << 11 | (w & 0x7ff)
            if op == 0x03:
                return "st.b", f"{REG[src1]},{s16(off):#x}({REG[src2]})"
            m = "st.l" if (off & 1) else "st.s"
            off &= 0xfffc if (off & 1) else 0xfffe
            return m, f"{REG[src1]},{s16(off):#x}({REG[src2]})"
        # FP fld/fst: FP reg in dest field; flags bit0=auto++, bit1:1=.l/0=.d, bit2=.q
        fl = w & 7
        sz = ".l" if (fl & 2) else (".q" if (fl & 4) else ".d")
        pp = "++" if (fl & 1) else ""
        m = ("fld" if op < 0x0a else "fst") + sz
        if op & 1:
            size = 4 if (fl & 2) else (16 if (fl & 4) else 8)
            ea = f"{s16(imm & (0x10000 - size)):#x}({REG[src2]}){pp}"
        else:
            ea = f"{REG[src1]}({REG[src2]}){pp}"
        if op < 0x0a:
            return m, f"{ea},{FREG[dest]}"
        return m, f"{FREG[dest]},{ea}"
    if op == 0x02:
        return "ixfr", f"{REG[src1]},{FREG[dest]}"
    if op == 0x0c:
        return "ld.c", f"{CTRL.get(src2,src2)},{REG[dest]}"
    if op == 0x0e:
        return "st.c", f"{REG[src1]},{CTRL.get(src2,src2)}"
    if op == 0x10:
        return "bri", f"{REG[src1]}"
    if op == 0x11:
        return "trap", f"{REG[src1]},{REG[src2]},{REG[dest]}"
    if op == 0x12:
        return dec_fp(w)
    if op == 0x13:
        # core escape: sub-op in low 5 bits (1=lock, 2=calli, 4=intovr, 7=unlock)
        sub = w & 0x1f
        if sub == 0x02:
            return "calli", f"{REG[src1]}"
        if sub == 0x01:
            return "lock", ""
        if sub == 0x07:
            return "unlock", ""
        if sub == 0x04:
            return "intovr", ""
        return "esc", f"{sub:#x}"
    if op in (0x14, 0x15, 0x16, 0x17):
        base = "btne" if op < 0x16 else "bte"
        broff = s16(((dest << 11) | (w & 0x7ff)) & 0xffff)
        tgt = addr + 4 + broff * 4
        # odd opcode = 5-bit-const form (i860 rule; e.g. `btne 0x0,r19` = 0x15)
        s1 = f"{src1:#x}" if op & 1 else f"{REG[src1]}"
        return base, f"{s1},{REG[src2]},{tgt:#010x}"
    if op in CTRLB:
        tgt = addr + 4 + s26(w & 0x03ffffff) * 4
        if CTRLB[op] in ("br", "call"):
            return CTRLB[op], f"{tgt:#010x}"
        return CTRLB[op], f"{tgt:#010x}"
    if op in (0x18, 0x19):
        off = s16(imm & 0xfff8)
        return "pfld.d", f"{off:#x}({REG[src2]}),{FREG[dest]}"
    if op == 0x2d:                    # bla isrc1,isrc2,sbroff (split offset like bte)
        off = s16((((w >> 16) & 0x1f) << 11 | (w & 0x7ff)) & 0xffff)
        return "bla", f"{REG[src1]},{REG[src2]},{addr + 4 + off*4:#010x}"
    if (op & 0x3e) in ARITH:
        m = ARITH[op & 0x3e]
        if op & 1:  # immediate form
            v = s16(imm)
            return m, f"{v:#x},{REG[src2]},{REG[dest]}"
        # register form; recognise pseudo-ops (mov/nop = shl r0)
        if m == "shl" and src1 == 0:
            if src2 == 0 and dest == 0:
                return "nop", ""
            return "mov", f"{REG[src2]},{REG[dest]}"
        if m == "or" and src1 == 0 and src2 == 0:
            return "mov", f"r0,{REG[dest]}"
        return m, f"{REG[src1]},{REG[src2]},{REG[dest]}"
    return ".word", f"{w:#010x}"

# ---------------- validation ----------------
def _norm_ops(s):
    """normalize an operand string for comparison: strip spaces, lower, and
    canonicalize hex/dec immediates."""
    import re
    s = s.lower().replace(' ', '')
    def canon(m):
        v = int(m.group(0), 0)
        return str(v)
    return re.sub(r'-?0x[0-9a-f]+|-?\b\d+\b', canon, s)

def validate():
    sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
    import derive860 as D
    base = r'C:\VWE\TeslaRel410\sda4\DPL3\VRENDER\AS860'
    # operand check only for ops whose operands are NOT link-time relocated
    OPCHECK = ('ld.', 'st.', 'fld.', 'fst.', 'adds', 'addu', 'subs', 'subu',
               'and', 'or', 'xor', 'shl', 'shr', 'shra', 'mov', 'ixfr', 'bri')
    for s in sorted(glob.glob(os.path.join(base, '*.S'))):
        o = s[:-2] + '.O'
        if not os.path.exists(o):
            continue
        labels, instrs = D.parse_S(s)
        text, syms = D.load_text(o)
        common = [k for k in labels if k in syms]
        aligned = not any(labels[k] != syms[k] for k in common)
        ok = tot = 0; oper_ok = oper_tot = 0; misses = []; omiss = []
        for (off, mnem, line) in instrs:
            if off + 4 > len(text):
                break
            w = struct.unpack_from('<I', text, off)[0]
            dm, dops = decode(w, off)
            tot += 1
            if dm.split('.')[0] == mnem.split('.')[0] or \
               dm.split('.')[0].lstrip('p') == mnem.split('.')[0] or \
               (mnem.startswith('pf') and dm.startswith('pf')):
                ok += 1
                # operand-level check where the source operands are literal
                if mnem.startswith(OPCHECK) and dm == mnem:
                    src_ops = line[len(mnem):].strip()
                    if not any(c.isalpha() and c not in 'rxf' for c in
                               src_ops.replace('r', '').replace('f', '')
                               .replace('x', '')):
                        oper_tot += 1
                        if _norm_ops(src_ops) == _norm_ops(dops):
                            oper_ok += 1
                        elif len(omiss) < 8:
                            omiss.append(f"@{off:#x} {mnem} src[{src_ops}] "
                                         f"dec[{dops}] w={w:#010x}")
            else:
                if len(misses) < 8:
                    misses.append(f"@{off:#x} src={mnem} dec={dm} w={w:#010x}")
        tag = os.path.basename(o)
        flag = "aligned" if aligned else "MISALIGNED"
        print(f"[{tag:14}] {flag:10} mnemonic {ok}/{tot}="
              f"{100*ok//max(tot,1)}%  operands {oper_ok}/{oper_tot}="
              f"{100*oper_ok//max(oper_tot,1)}%")
        if aligned:
            for m in misses:
                print("      mn-miss ", m)
            for m in omiss:
                print("      op-miss ", m)

def disasm_range(text, start, count, base=0x0c):
    """Disassemble `count` words of `text` from byte offset `start`.
    `base` = file offset where .text begins (VREND.MNG = 0x0c), so printed
    addresses are file/memory offsets."""
    out = []
    for i in range(count):
        off = start + i*4
        if off + 4 > len(text): break
        w = struct.unpack_from('<I', text, off)[0]
        # decode with the PRINTED address so branch targets match the listing
        m, ops = decode(w, base+off)
        out.append(f"  {base+off:#08x}: {w:08x}  {m:<10} {ops}")
    return "\n".join(out)

def load_mng(path):
    d = open(path, 'rb').read()
    tsize = struct.unpack_from('<I', d, 0)[0]
    return d[0x0c:0x0c+tsize]

if __name__ == '__main__':
    if '--validate' in sys.argv:
        validate()
    elif '--list' in sys.argv:
        # dis860.py --list <mng> <hexoffset> <count>
        a = [x for x in sys.argv if not x.startswith('--')][1:]
        mng = a[0]; start = int(a[1], 0); count = int(a[2]) if len(a) > 2 else 32
        print(disasm_range(load_mng(mng), start, count))
    else:
        print("usage: dis860.py --validate | --list <VREND.MNG> <off> <count>")