I’m currently making a buffer compression algorithm to store builds in datastore more easily but it is very slow on big buffers and I’ve found this function which takes 50% of the decode computation time, how could I optimize it ?
function BufferOperations.ReadBinaryStringBuffer(b: buffer)
-- Variables
local bufferLen = buffer.len(b)
local lastByteBits = buffer.readu8(b, bufferLen - 1)
local outputString = {}
-- Main
for byteIndex = 0, bufferLen - 2 do
local byte = buffer.readu8(b, byteIndex)
local bitsToRead = 8
if byteIndex == bufferLen - 2 then
bitsToRead = lastByteBits
end
for bitPos = 0, bitsToRead - 1 do
table.insert(outputString, bit32.extract(byte, bitPos))
end
end
-- Returning
return table.concat(outputString)
end
Something that come to mind is using buffer.readu32 calls for the largest segment of the string, so you read multiple bytes at once and do fewer iterations. Another idea is to manually maintain and pass the size/length of the buffer, instead of doing buffer.len calls and scanning over the entire thing every time. You can also use table.create(n) for the outputString, since, if I’m not mistaken, the size of the table can be precomputed(it doesn’t rely on complex behavior).
The opposite, actually; due to Roblox’s dumb tendency of making everything a method and probably Luau → Instance API bridge overhead, Luau implementation will be MUCH MUCH faster thanks to native code generation that will actually work and be compiled in this case.
Yes, it will generate native code in this case; I fact-checked it
Nope, sadly the compiler is too stupid; it won’t.
It has NEWTABLE and DUPTABLE opcodes that always go to waste. Even if the table.create(n) argument passed is static, the compiler won’t generate the NEWTABLE opcode instead, and it’s such a shame.
Like it makes 100% sense for table.create(100) to be NEWTABLE R0 100 0 but nope, the compiler is an idiot.
