a microbenchmark comparing different loop types, and different work size per-iteration

loop_benchmark.lua

--config
local replicas = 3
local test_duration = 0.5
local single_test_size = 1e3

--per-iteration operation

function test_operation(a, b, work_size)
	for _ = 1, work_size do
		local s = tostring(a + b)
	end
end

--different loop constructs

function use_pairs(a, work_size)
	for i, v in pairs(a) do
		test_operation(i, v, work_size)
	end
end

function use_ipairs(a, work_size)
	for i, v in ipairs(a) do
		test_operation(i, v, work_size)
	end
end

function use_numeric(a, work_size)
	for i = 1, #a do
		test_operation(i, a[i], work_size)
	end
end

--test setup

function test(work_size)
	--init array
	local a = {}
	for i = 1, single_test_size do
		a[i] = i * i
	end

	local results = {}
	for cfg_i, cfg in ipairs({
		{"ipairs",  use_ipairs},
		{"numeric", use_numeric},
		{"pairs",   use_pairs},
	}) do
		local name, f = unpack(cfg)

		--count loops
		local loops = 0
		local start = love.timer.getTime()
		while love.timer.getTime() - start < test_duration do
			f(a, work_size)
			loops = loops + 1
		end

		--measure the actual time taken
		local time_taken = love.timer.getTime() - start

		--compare to first result
		table.insert(results, loops)
		local ratio = results[1] / results[#results]

		--output
		print(string.format("%10s | %8d loops | %12.3f loops/s | ratio % 8.3fx", name, loops, loops / time_taken, ratio))

		--ready for next test
		collectgarbage("collect")
	end
end

function love.load()
	for _, work_size in ipairs({
		0, 1, 2, 4, 8
	}) do
		print("-------------------------------")
		print("comparing with work size: "..work_size)
		for i = 1, replicas do
			print("replica", i)
			test(work_size)
		end
	end
	love.event.quit()
end

loop_benchmark_example_output.txt

Trial Run Results

i5-4690 @ 3.50GHz, 16GB RAM
Windows 10 x64
Love 11.2

-------------------------------
comparing with work size: 0
replica	1
    ipairs |   588479 loops |  1176957.656 loops/s | ratio    1.000x
   numeric |   849657 loops |  1699311.514 loops/s | ratio    0.693x
     pairs |    27511 loops |    55021.920 loops/s | ratio   21.391x
replica	2
    ipairs |   589456 loops |  1178910.965 loops/s | ratio    1.000x
   numeric |   852279 loops |  1704556.504 loops/s | ratio    0.692x
     pairs |    27606 loops |    55211.273 loops/s | ratio   21.352x
replica	3
    ipairs |   587989 loops |  1175976.968 loops/s | ratio    1.000x
   numeric |   853027 loops |  1706052.502 loops/s | ratio    0.689x
     pairs |    27566 loops |    55131.564 loops/s | ratio   21.330x
-------------------------------
comparing with work size: 1
replica	1
    ipairs |    65388 loops |   130774.202 loops/s | ratio    1.000x
   numeric |    66449 loops |   132897.495 loops/s | ratio    0.984x
     pairs |      771 loops |     1540.362 loops/s | ratio   84.809x
replica	2
    ipairs |    63015 loops |   126029.299 loops/s | ratio    1.000x
   numeric |    66798 loops |   133594.163 loops/s | ratio    0.943x
     pairs |      771 loops |     1540.623 loops/s | ratio   81.732x
replica	3
    ipairs |    64957 loops |   129913.354 loops/s | ratio    1.000x
   numeric |    67794 loops |   135586.294 loops/s | ratio    0.958x
     pairs |      783 loops |     1564.903 loops/s | ratio   82.959x
-------------------------------
comparing with work size: 2
replica	1
    ipairs |      781 loops |     1560.757 loops/s | ratio    1.000x
   numeric |      785 loops |     1569.734 loops/s | ratio    0.995x
     pairs |      417 loops |      833.122 loops/s | ratio    1.873x
replica	2
    ipairs |      774 loops |     1547.051 loops/s | ratio    1.000x
   numeric |      781 loops |     1561.610 loops/s | ratio    0.991x
     pairs |      415 loops |      829.380 loops/s | ratio    1.865x
replica	3
    ipairs |      773 loops |     1545.269 loops/s | ratio    1.000x
   numeric |      751 loops |     1499.788 loops/s | ratio    1.029x
     pairs |      406 loops |      810.658 loops/s | ratio    1.904x
-------------------------------
comparing with work size: 4
replica	1
    ipairs |      284 loops |      567.361 loops/s | ratio    1.000x
   numeric |      285 loops |      568.091 loops/s | ratio    0.996x
     pairs |      218 loops |      435.634 loops/s | ratio    1.303x
replica	2
    ipairs |      286 loops |      571.570 loops/s | ratio    1.000x
   numeric |      287 loops |      572.868 loops/s | ratio    0.997x
     pairs |      217 loops |      433.892 loops/s | ratio    1.318x
replica	3
    ipairs |      287 loops |      572.192 loops/s | ratio    1.000x
   numeric |      287 loops |      573.741 loops/s | ratio    1.000x
     pairs |      217 loops |      433.250 loops/s | ratio    1.323x
-------------------------------
comparing with work size: 8
replica	1
    ipairs |      127 loops |      252.785 loops/s | ratio    1.000x
   numeric |      128 loops |      254.925 loops/s | ratio    0.992x
     pairs |      112 loops |      223.323 loops/s | ratio    1.134x
replica	2
    ipairs |      129 loops |      256.056 loops/s | ratio    1.000x
   numeric |      129 loops |      257.458 loops/s | ratio    1.000x
     pairs |      112 loops |      223.185 loops/s | ratio    1.152x
replica	3
    ipairs |      128 loops |      255.384 loops/s | ratio    1.000x
   numeric |      129 loops |      257.310 loops/s | ratio    0.992x
     pairs |      113 loops |      225.747 loops/s | ratio    1.133x