VoxelEngine/res/modules/bit_converter.lua
2025-01-21 02:30:59 +03:00

367 lines
9.7 KiB
Lua

local bit_converter = { }
local MAX_UINT16 = 65535
local MIN_UINT16 = 0
local MAX_UINT32 = 4294967295
local MIN_UINT32 = 0
local MAX_INT16 = 32767
local MIN_INT16 = -32768
local MAX_INT32 = 2147483647
local MIN_INT32 = -2147483648
local MAX_INT64 = 9223372036854775807
local MIN_INT64 = -9223372036854775808
local function maskHighBytes(num)
return bit.band(num, 0xFF)
end
local function reverse(tbl)
for i=1, math.floor(#tbl / 2) do
local tmp = tbl[i]
tbl[i] = tbl[#tbl - i + 1]
tbl[#tbl - i + 1] = tmp
end
return tbl
end
local orders = { "LE", "BE" }
local fromLEConvertors =
{
LE = function(bytes) return bytes end,
BE = function(bytes) return reverse(bytes) end
}
local toLEConvertors =
{
LE = function(bytes) return bytes end,
BE = function(bytes) return reverse(bytes) end
}
bit_converter.default_order = "BE"
local function fromLE(bytes, orderTo)
if orderTo then
bit_converter.validate_order(orderTo)
return fromLEConvertors[orderTo](bytes)
else return bytes end
end
local function toLE(bytes, orderFrom)
if orderFrom then
bit_converter.validate_order(orderFrom)
return toLEConvertors[orderFrom](bytes)
else return bytes end
end
function bit_converter.validate_order(order)
if not bit_converter.is_valid_order(order) then
error("invalid order: "..order)
end
end
function bit_converter.is_valid_order(order) return table.has(orders, order) end
function bit_converter.string_to_bytes(str)
local bytes = { }
local len = string.len(str)
local lenBytes = bit_converter.uint16_to_bytes(len)
for i = 1, #lenBytes do
bytes[i] = lenBytes[i]
end
for i = 1, len do
bytes[#bytes + 1] = string.byte(string.sub(str, i, i))
end
return bytes
end
function bit_converter.bool_to_byte(bool)
return bool and 1 or 0
end
-- Credits to Iryont <https://github.com/iryont/lua-struct>
local function floatOrDoubleToBytes(val, opt)
local sign = 0
if val < 0 then
sign = 1
val = -val
end
local mantissa, exponent = math.frexp(val)
if val == 0 then
mantissa = 0
exponent = 0
else
mantissa = (mantissa * 2 - 1) * math.ldexp(0.5, (opt == 'd') and 53 or 24)
exponent = exponent + ((opt == 'd') and 1022 or 126)
end
local bytes = {}
if opt == 'd' then
val = mantissa
for i = 1, 6 do
bytes[#bytes + 1] = math.floor(val) % (2 ^ 8)
val = math.floor(val / (2 ^ 8))
end
else
bytes[#bytes + 1] = math.floor(mantissa) % (2 ^ 8)
val = math.floor(mantissa / (2 ^ 8))
bytes[#bytes + 1] = math.floor(val) % (2 ^ 8)
val = math.floor(val / (2 ^ 8))
end
bytes[#bytes + 1] = math.floor(exponent * ((opt == 'd') and 16 or 128) + val) % (2 ^ 8)
val = math.floor((exponent * ((opt == 'd') and 16 or 128) + val) / (2 ^ 8))
bytes[#bytes + 1] = math.floor(sign * 128 + val) % (2 ^ 8)
val = math.floor((sign * 128 + val) / (2 ^ 8))
return bytes
end
local function bytesToFloatOrDouble(bytes, opt)
local n = (opt == 'd') and 8 or 4
local sign = 1
local mantissa = bytes[n - 1] % ((opt == 'd') and 16 or 128)
for i = n - 2, 1, -1 do
mantissa = mantissa * (2 ^ 8) + bytes[i]
end
if bytes[n] > 127 then
sign = -1
end
local exponent = (bytes[n] % 128) * ((opt == 'd') and 16 or 2) + math.floor(bytes[n - 1] / ((opt == 'd') and 16 or 128))
if exponent == 0 then
return 0.0
else
mantissa = (math.ldexp(mantissa, (opt == 'd') and -52 or -23) + 1) * sign
return math.ldexp(mantissa, exponent - ((opt == 'd') and 1023 or 127))
end
end
--
function bit_converter.float32_to_bytes(float, order)
return fromLE(floatOrDoubleToBytes(float, 'f'), order)
end
function bit_converter.float64_to_bytes(float, order)
return fromLE(floatOrDoubleToBytes(float, 'd'), order)
end
function bit_converter.single_to_bytes(float, order)
on_deprecated_call("bit_converter.float_to_bytes", "bit_converter.float32_to_bytes")
return bit_converter.float32_to_bytes(bytes, order)
end
function bit_converter.double_to_bytes(double, order)
on_deprecated_call("bit_converter.double_to_bytes", "bit_converter.float64_to_bytes")
return bit_converter.float64_to_bytes(bytes, order)
end
local function uint32ToBytes(int, order)
return fromLE({
maskHighBytes(int),
maskHighBytes(bit.rshift(int, 8)),
maskHighBytes(bit.rshift(int, 16)),
maskHighBytes(bit.rshift(int, 24))
}, order)
end
local function uint16ToBytes(int, order)
return fromLE({
maskHighBytes(int),
maskHighBytes(bit.rshift(int, 8))
}, order)
end
function bit_converter.uint32_to_bytes(int, order)
if int > MAX_UINT32 or int < MIN_UINT32 then
error("invalid uint32")
end
return uint32ToBytes(int, order)
end
function bit_converter.uint16_to_bytes(int, order)
if int > MAX_UINT16 or int < MIN_UINT16 then
error("invalid uint16")
end
return uint16ToBytes(int, order)
end
function bit_converter.int64_to_bytes(int, order)
if int > MAX_INT64 or int < MIN_INT64 then
error("invalid int64")
end
return fromLE({
maskHighBytes(int),
maskHighBytes(bit.rshift(int, 8)),
maskHighBytes(bit.rshift(int, 16)),
maskHighBytes(bit.rshift(int, 24)),
maskHighBytes(bit.rshift(int, 32)),
maskHighBytes(bit.rshift(int, 40)),
maskHighBytes(bit.rshift(int, 48)),
maskHighBytes(bit.rshift(int, 56))
}, order)
end
function bit_converter.int32_to_bytes(int, order)
on_deprecated_call("bit_converter.int32_to_bytes", "bit_converter.sint32_to_bytes")
if int > MAX_INT32 or int < MIN_INT32 then
error("invalid int32")
end
return uint32ToBytes(int + MAX_INT32, order)
end
function bit_converter.int16_to_bytes(int, order)
on_deprecated_call("bit_converter.int32_to_bytes", "bit_converter.sint16_to_bytes")
if int > MAX_INT16 or int < MIN_INT16 then
error("invalid int16")
end
return uint16ToBytes(int + MAX_INT16, order)
end
function bit_converter.sint32_to_bytes(int, order)
if int > MAX_INT32 or int < MIN_INT32 then
error("invalid sint32")
end
return uint32ToBytes(int + MAX_UINT32 + 1, order)
end
function bit_converter.sint16_to_bytes(int, order)
if int > MAX_INT16 or int < MIN_INT16 then
error("invalid sint16")
end
return uint16ToBytes(int + MAX_UINT16 + 1, order)
end
function bit_converter.bytes_to_float32(bytes, order)
return bytesToFloatOrDouble(toLE(bytes, order), 'f')
end
function bit_converter.bytes_to_float64(bytes, order)
return bytesToFloatOrDouble(toLE(bytes, order), 'd')
end
function bit_converter.bytes_to_single(bytes, order)
on_deprecated_call("bit_converter.bytes_to_single", "bit_converter.bytes_to_float32")
return bit_converter.bytes_to_float32(bytes, order)
end
function bit_converter.bytes_to_double(bytes, order)
on_deprecated_call("bit_converter.bytes_to_double", "bit_converter.bytes_to_float64")
return bit_converter.bytes_to_float64(bytes, order)
end
function bit_converter.bytes_to_string(bytes, order)
local len = bit_converter.bytes_to_uint16({ bytes[1], bytes[2] })
local str = ""
for i = 1, len do
str = str..string.char(bytes[i + 2])
end
return str
end
function bit_converter.byte_to_bool(byte)
return byte ~= 0
end
function bit_converter.bytes_to_uint32(bytes, order)
if #bytes < 4 then
error("eof")
end
bytes = toLE(bytes, order)
return
bit.bor(
bit.bor(
bit.bor(
bytes[1],
bit.lshift(bytes[2], 8)),
bit.lshift(bytes[3], 16)),
bit.lshift(bytes[4], 24))
end
function bit_converter.bytes_to_uint16(bytes, order)
if #bytes < 2 then
error("eof")
end
bytes = toLE(bytes, order)
return
bit.bor(
bit.lshift(bytes[2], 8),
bytes[1], 0)
end
function bit_converter.bytes_to_int64(bytes, order)
if #bytes < 8 then
error("eof")
end
bytes = toLE(bytes, order)
return
bit.bor(
bit.bor(
bit.bor(
bit.bor(
bit.bor(
bit.bor(
bit.bor(
bit.lshift(bytes[8], 56),
bit.lshift(bytes[7], 48)),
bit.lshift(bytes[6], 40)),
bit.lshift(bytes[5], 32)),
bit.lshift(bytes[4], 24)),
bit.lshift(bit.band(bytes[3], 0xFF), 16)),
bit.lshift(bit.band(bytes[2], 0xFF), 8)),bit.band(bytes[1], 0xFF))
end
function bit_converter.bytes_to_int32(bytes, order)
on_deprecated_call("bit_converter.bytes_to_int32", "bit_converter.bytes_to_sint32")
return bit_converter.bytes_to_uint32(bytes, order) - MAX_INT32
end
function bit_converter.bytes_to_int16(bytes, order)
on_deprecated_call("bit_converter.bytes_to_int16", "bit_converter.bytes_to_sint16")
return bit_converter.bytes_to_uint16(bytes, order) - MAX_INT16
end
function bit_converter.bytes_to_sint32(bytes, order)
local num = bit_converter.bytes_to_uint32(bytes, order)
return MIN_INT32 * (bit.band(MAX_INT32 + 1, num) ~= 0 and 1 or 0) + bit.band(MAX_INT32, num)
end
function bit_converter.bytes_to_sint16(bytes, order)
local num = bit_converter.bytes_to_uint16(bytes, order)
return MIN_INT16 * (bit.band(MAX_INT16 + 1, num) ~= 0 and 1 or 0) + bit.band(MAX_INT16, num)
end
return bit_converter