Benchmark of C callbacks in LuaJIT 2.0.0-beta10

luajit_benchmark_callback.lua

local ffi = require"ffi"
jit.on()

ffi.cdef[[
    typedef int (*getval_func)();
    
    /* imagine these are C functions */
    typedef void (*sum_push_func)(int* total, getval_func getval);
    typedef void (*sum_pull_func)(int* total, int val);
]]

local sum_push = ffi.cast("sum_push_func", function(total_ptr, getval_ptr)
    total_ptr[0] = total_ptr[0]+getval_ptr()
end)
local sum_pull = ffi.cast("sum_pull_func", function(total_ptr, val)
    total_ptr[0] = total_ptr[0]+val
end)

local getval = function()
    return 6
end
local getval_ptr = ffi.cast("getval_func", getval)


function test_push(n)
    local total_ptr = ffi.new("int[1]", 0)
    for i = 1, n do
        sum_push(total_ptr, getval_ptr)
    end
    return total_ptr[0]
end

function test_pull(n)
    local total_ptr = ffi.new("int[1]", 0)
    for i = 1, n do
        sum_pull(total_ptr, getval())
    end
    return total_ptr[0]
end

local gettime = os.clock
local fmt = "%-10s %-16s %-16s %-16s %-16s"
local jit_options = {jit.status()} local jit_on = table.remove(jit_options, 1)
print(jit.version.. " ("..jit.os.." "..jit.arch..")")
print("JIT: "..(jit_on and "ON" or "OFF").." "..table.concat(jit_options, " "))
print(fmt:format("n", "Push Time", "Pull Time", "Push Mem", "Pull Mem"))
function math.round(n, p) return math.floor(n*p)/p end
local trials = 3
for i = 8, 24, 4 do
    local n = 2^i
    
    local time_total_push, time_total_pull = 0, 0
    local memu_total_push, memu_total_pull = 0, 0
    local res_push, res_pull
    for i = 1, trials do
        collectgarbage("collect")
        collectgarbage("collect")
        local time_push = os.clock()
        res_push = test_push(n)
        time_push = os.clock()-time_push
        time_total_push = time_total_push+time_push
        memu_total_push = memu_total_push+collectgarbage("count")
        
        collectgarbage("collect")
        collectgarbage("collect")
        local time_pull = os.clock()
        res_pull = test_pull(n)
        time_pull = os.clock()-time_pull
        time_total_pull = time_total_pull+time_pull
        memu_total_pull = memu_total_pull+collectgarbage("count")
    end
    
    print(fmt:format(
        n,
        math.round(time_total_push/trials, 10^6),
        math.round(time_total_pull/trials, 10^6),
        math.round(memu_total_push/trials, 10^6),
        math.round(memu_total_pull/trials, 10^6)
    ))
end

--[[
LuaJIT 2.0.0-beta10 (Windows x64)
JIT: ON CMOV SSE2 SSE3 SSE4.1 fold cse dce fwd dse narrow loop abc sink fuse
n          Push Time        Pull Time        Push Mem         Pull Mem
256        0                0                63.247721        59.346679
4096       0.002666         0.000999         183.346679       119.346679
65536      0.036666         0.015333         2103.346679      1079.346679
1048576    0.620666         0.272            32823.346679     16439.346679
16777216   9.439666         4.227666         524343.346679    262199.346679
]]

LuaJIT 2.0.1 (OSX x64)
JIT: ON CMOV SSE2 SSE3 SSE4.1 fold cse dce fwd dse narrow loop abc sink fuse
n          Push Time        Pull Time        Push Mem         Pull Mem
256        0.000391         0.000156         47.843424        43.942382
4096       0.002765         0.000862         167.942382       103.942382
65536      0.033591         0.013966         2087.942382      1063.942382
1048576    0.531986         0.22839          32807.942382     16423.942382
16777216   8.680506         3.688063         524327.942382    262183.942382

LuaJIT 2.1.0-beta3 (Windows x64)
JIT: ON SSE2 SSE3 SSE4.1 AMD BMI2 fold cse dce fwd dse narrow loop abc sink fuse
n          Push Time        Pull Time        Push Mem         Pull Mem
256        0.000333         0                48.187174        44.293945
4096       0.002333         0.000333         69.856445        70.450195
65536      0.032666         0.014            66.668945        81.168945
1048576    0.521            0.217            64.981445        70.575195
16777216   8.422            3.470333         87.293945        83.168945

i3-3220 (3.3GHz) + 1600MHz DDR3, no foreground processes:

LuaJIT 2.1.0-beta3 (Linux x64)
JIT: ON SSE2 SSE3 SSE4.1 fold cse dce fwd dse narrow loop abc sink fuse
n          Push Time        Pull Time        Push Mem         Pull Mem        
256        0.000391         0.000156         48.342447        44.449218       
4096       0.003357         0.001251         69.699218        70.605468       
65536      0.031069         0.0142           64.074218        76.636718       
1048576    0.496462         0.228811         72.824218        76.152343       
16777216   7.958449         3.643323         64.699218        68.402343       

LuaJIT 2.1.1710088188 (Windows x64)
JIT: ON SSE3 SSE4.1 fold cse dce fwd dse narrow loop abc sink fuse
n          Push Time        Pull Time        Push Mem         Pull Mem
256        0.000666         0.000333         65.730794        59.944335
4096       0.006            0.002666         113.381835       84.014648
65536      0.058333         0.026            72.881835        73.561523
1048576    0.903333         0.418999         88.631835        104.100585
16777216   14.506666        6.770666         75.131835        89.217773