Log In  

Does what it says on the tin. Lets you see what's eating up your CPU time. If you make an extension to this, I'd love to see it.

API:

profile.enabled(detailed,cpu) Turns on and off profiling tools.
detailed: Whether or not to display results of profile calls.
cpu: Whether or not to display total CPU usage.

profile(name,linger) Starts or stops a profile. Accumulates between profile.draw calls. Extremely cheap when profiling is disabled.
name: Arbitrary display name indicating which profile to start or stop.
linger: Whether or not the profile should linger even if the profile is never called between draws.

profile.draw() Draws profile and CPU information to the screen if they are enabled. Extremely cheap when profiling is disabled.

profile.clear_lingers() Clears any lingering profile information.

Example:

include("profiler.lua")

profile.enabled(true,true)

local frames = 0

function _update()
  profile("_update")
  frames = frames+1
  profile("_update")
end

function _draw()
  profile("_draw")
  if frames%2 == 0 then
    profile("displaying frame count",true)
    print(frames,80,0,7)
    profile("displaying frame count",true)
  end
  profile.draw()
  profile("_draw")
end

Full script:

--[[pod_format="raw",created="2024-04-09 22:52:04",modified="2024-04-11 17:26:16",revision=1003]]
-- abledbody's profiler v1.1

local function do_nothing() end

-- The metatable here is to make profile() possible.
-- Why use a table at all? Because otherwise lua will try to cache the function call,
-- which by default is do_nothing.
local profile_meta = {__call = do_nothing}
profile = {draw = do_nothing}
setmetatable(profile,profile_meta)

local running = {} -- All incomplete profiles
 -- All complete profiles. Note that if the profiles haven't been drawn yet, it will
 -- not be cleared, and further profiles of the same name will add to the usage metric.
local profiles = {}
-- All completed lingering profiles. These are never automatically cleared.
local lingers = {}

-- start_profile, stop_profile, and stop_linger are all internal functions,
-- serving as paths for _profile to take. Lingers share start_profile.
local function start_profile(name,linger)
    local source = profiles[name]
    running[name] = {
        linger = linger,
    }
    local active = running[name]
    active.start = stat(1) --Delaying CPU usage grab until the last possible second.
end

local function stop_profile(name,active,delta)
    local profile = profiles[name]
    if profile then
        profile.time = delta+profile.time
    else
        profiles[name] = {
            time = delta,
            name = name,
        }
        add(profiles,profiles[name])
    end
end

local function stop_linger(name,active,delta)
    local profile = lingers[name]
    if profile then
        profile.time = profile.this_frame and delta+profile.time or delta
        profile.this_frame = true
    else
        lingers[name] = {
            time = delta,
            this_frame = true,
        }
    end
end

-- The main functionality lives here.
-- Takes in the name of what you're profiling, and whether or not to
-- make the profile linger.
local function _profile(_,name,linger)
    local t = stat(1)
    local active = running[name]
    if active then
        local delta = t-active.start

        if active.linger then stop_linger(name,active,delta)
        else stop_profile(name,active,delta) end

        running[name] = nil
    else
        start_profile(name,linger)
    end
end

-- Clears all lingering profiles.
function profile.clear_lingers()
    lingers = {}
end

local function draw_cpu()
    print("cpu:"..string.sub(stat(1)*100,1,5).."%",1,1,7)
end

-- This draws the profiles, and then resets everything for the next frame.
-- If it is not called, usage metrics will accumulate.
-- Lingering profiles are always displayed after persistent profiles.
local function display_profiles()
    local i = 1
    for prof in all(profiles) do
        local usage = string.sub(prof.time*100,1,5).."%"
        local to_print = prof.name..":"..usage
        print(to_print,1,1+i*9,7)
        i = i+1
    end
    for name,prof in pairs(lingers) do
        local usage = string.sub(prof.time*100,1,5).."%"
        local to_print = name..(prof.this_frame and "[X]:" or "[ ]:")..usage
        print(to_print,1,1+i*9,7)
        prof.this_frame = false
        i = i+1
    end
    profiles = {}
end

local function display_both()
    draw_cpu()
    display_profiles()
end

-- This swaps out function calls depending on whether or not you want to have
-- profiling. This is to make it as much as possible so that you don't have to
-- think about cleaning up profile calls for efficiency.
-- The first boolean is for detailed profiling, the second is for CPU usage.
function profile.enabled(detailed,cpu)
    profile_meta.__call = detailed and _profile or do_nothing
    profile.draw = detailed and (cpu and display_both or display_profiles)
        or (cpu and draw_cpu or do_nothing)
end

P#146421 2024-04-12 04:30


[Please log in to post a comment]