Performance of b64 encoding/decoding

[vincentjames501]

I'd like to say this is a great pure Lua approach to doing b64 encoding/decoding. We really need this to do b32 decoding and b64 url safe decoding and finding that easily in other modules is a challenge!

When compared to other modules (such as https://github.com/iskolbin/lbase64), the encoding/decoding performance looks like it could use some attention. I don't do much Lua so not sure how much help I would be here. Here is a benchmark on my M2 MBP:

basexx = require "basexx"
base64 = require "base64"

local units = {
    ['seconds'] = 1,
    ['milliseconds'] = 1000,
    ['microseconds'] = 1000000,
    ['nanoseconds'] = 1000000000
}

function benchmark(unit, decPlaces, n, f, ...)
    local elapsed = 0
    local multiplier = units[unit]
    for i = 1, n do
        local now = os.clock()
        f(...)
        elapsed = elapsed + (os.clock() - now)
    end
    print(string.format('Benchmark results: %d function calls | %.'.. decPlaces ..'f %s elapsed | %.'.. decPlaces ..'f %s avg execution time.', n, elapsed * multiplier, unit, (elapsed / n) * multiplier, unit))
end

print("Benchmarking base64.decode...")

benchmark('milliseconds', 5, 10000, base64.decode, "TWFuIGlzIGRpc3Rpbmd1aXNoZWQsIG5vdCBvbmx5IGJ5IGhpcyByZWFzb24sIGJ1dCBieSB0aGlzIHNpbmd1bGFyIHBhc3Npb24gZnJvbSBvdGhlciBhbmltYWxzLCB3aGljaCBpcyBhIGx1c3Qgb2YgdGhlIG1pbmQsIHRoYXQgYnkgYSBwZXJzZXZlcmFuY2Ugb2YgZGVsaWdodCBpbiB0aGUgY29udGludWVkIGFuZCBpbmRlZmF0aWdhYmxlIGdlbmVyYXRpb24gb2Yga25vd2xlZGdlLCBleGNlZWRzIHRoZSBzaG9ydCB2ZWhlbWVuY2Ugb2YgYW55IGNhcm5hbCBwbGVhc3VyZS4=")

print("Benchmarking basexx.from_base64...")

benchmark('milliseconds', 5, 10000, basexx.from_base64, "TWFuIGlzIGRpc3Rpbmd1aXNoZWQsIG5vdCBvbmx5IGJ5IGhpcyByZWFzb24sIGJ1dCBieSB0aGlzIHNpbmd1bGFyIHBhc3Npb24gZnJvbSBvdGhlciBhbmltYWxzLCB3aGljaCBpcyBhIGx1c3Qgb2YgdGhlIG1pbmQsIHRoYXQgYnkgYSBwZXJzZXZlcmFuY2Ugb2YgZGVsaWdodCBpbiB0aGUgY29udGludWVkIGFuZCBpbmRlZmF0aWdhYmxlIGdlbmVyYXRpb24gb2Yga25vd2xlZGdlLCBleGNlZWRzIHRoZSBzaG9ydCB2ZWhlbWVuY2Ugb2YgYW55IGNhcm5hbCBwbGVhc3VyZS4=")

print("Benchmarking base64.encode...")

benchmark('milliseconds', 5, 10000, base64.encode, "Man is distinguished, not only by his reason, but by this singular passion from other animals, which is a lust of the mind, that by a perseverance of delight in the continued and indefatigable generation of knowledge, exceeds the short vehemence of any carnal pleasure.")

print("Benchmarking basexx.to_base64...")

benchmark('milliseconds', 5, 10000, basexx.to_base64, "Man is distinguished, not only by his reason, but by this singular passion from other animals, which is a lust of the mind, that by a perseverance of delight in the continued and indefatigable generation of knowledge, exceeds the short vehemence of any carnal pleasure.")

Results:

Benchmarking base64.decode...
Benchmark results: 10000 function calls | 259.39300 milliseconds elapsed | 0.02594 milliseconds avg execution time.
Benchmarking basexx.from_base64...
Benchmark results: 10000 function calls | 5682.15300 milliseconds elapsed | 0.56822 milliseconds avg execution time.
Benchmarking base64.encode...
Benchmark results: 10000 function calls | 219.74300 milliseconds elapsed | 0.02197 milliseconds avg execution time.
Benchmarking basexx.to_base64...
Benchmark results: 10000 function calls | 5304.85800 milliseconds elapsed | 0.53049 milliseconds avg execution time.

So it's around 20-25x slower. Any thoughts on what can be done to improve the performance?

[aiq]

Thanks for the feedback.

The implementation had allways room for performance improvements [1]..
But it was never an issue for me to improve it.

I can see that a implementation that uses the native bit operators like lbase64 has a significant boost in performance.

You are welcome to create a push request with improvements. I personally will not find the time in the near future to make the changes by myself.