[joaquimandrade]

Originally Posted by SchlumPF* never used pev_valid(id) == 2 but i guess it is not really needed. if you set some ents pdata you should already know whether it has pdata. but it may be usefull if you loop through different kinds of ents (classnames)

I know what it is but also never needed to check if "equals 2".

Originally Posted by SchlumPF* a new question which came to my mind is there any difference in client_preThink( plr ) and register_forward( FM_PlayerPreThink, "fwdPlayerPreThink" ) ?

I guess that there isn't any difference.

[SchlumPF*]

i just answered your question, i think that the function itself is just usefull if you need to loop through different types of ents.

[Hawk552]

I'm not sure but I'd guess that client_PreThink() becomes much faster once there's more than one plugin using it, since it's a multiforward.

[hleV]

You should do like 3-5 tests to compare each FM's & Engine's function type speed.

Like get_pdata_int() 3-5 times & Engine's way that does the same 3-5 times. That would make it clear for everyone which way is better.

[xPaw]

Engine/Fakemeta/HamSandwich PreThink test.

date: Mon Mar 30 16:38:56 2009 map: de_dust2
type |                             name |      calls | time / min / max
-------------------------------------------------------------------
   n |                  register_plugin |          1 | 0.000004 / 0.000004 / 0.000004
   n |                 register_forward |          1 | 0.000009 / 0.000009 / 0.000009
   n |                      RegisterHam |          1 | 0.000091 / 0.000091 / 0.000091
   n |                         set_task |          1 | 0.000006 / 0.000006 / 0.000006
   n |                          log_amx |          3 | 0.007320 / 0.001663 / 0.003362
   n |                       server_cmd |          1 | 0.000014 / 0.000014 / 0.000014
   p |                  client_PreThink |       1688 | 0.004612 / 0.000001 / 0.000422
   p |              fwdFMPlayerPreThink |       1688 | 0.002990 / 0.000001 / 0.000536
   p |             fwdHamPlayerPreThink |       1688 | 0.002635 / 0.000001 / 0.000255
   p |                 fnShutDownServer |          1 | 0.000013 / 0.000013 / 0.000013
   p |                      plugin_init |          1 | 0.000006 / 0.000006 / 0.000006
6 natives, 1 public callbacks, 6 function calls were not executed.

[Hawk552]

Originally Posted by xPaw Engine/Fakemeta/HamSandwich PreThink test. date: Mon Mar 30 16:38:56 2009 map: de_dust2 type | name | calls | time / min / max ------------------------------------------------------------------- n | register_plugin | 1 | 0.000004 / 0.000004 / 0.000004 n | register_forward | 1 | 0.000009 / 0.000009 / 0.000009 n | RegisterHam | 1 | 0.000091 / 0.000091 / 0.000091 n | set_task | 1 | 0.000006 / 0.000006 / 0.000006 n | log_amx | 3 | 0.007320 / 0.001663 / 0.003362 n | server_cmd | 1 | 0.000014 / 0.000014 / 0.000014 p | client_PreThink | 1688 | 0.004612 / 0.000001 / 0.000422 p | fwdFMPlayerPreThink | 1688 | 0.002990 / 0.000001 / 0.000536 p | fwdHamPlayerPreThink | 1688 | 0.002635 / 0.000001 / 0.000255 p | fnShutDownServer | 1 | 0.000013 / 0.000013 / 0.000013 p | plugin_init | 1 | 0.000006 / 0.000006 / 0.000006 6 natives, 1 public callbacks, 6 function calls were not executed.

I don't think that's a fair test since you're basically benchmarking CreateOneForward against CreateMultiForward. If you create a bunch of plugins which use client_PreThink, it will likely be faster.

The other problem is that you're not benchmarking how long it takes for the modules to find and call each function. You can't really profile forwards. As far as I know, it can only benchmark stuff done in the VM layer.

A good test would be to benchmark how long it takes between PreThink and PostThink. You could probably draw a more accurate conclusion from that.

[joaquimandrade]

I have two questions:

How can increment a variable inside a forward be less or more expensive depending on which forward?

What is the point of profiling forwards?

Anyway, i've made more neutral profiling for you. Here it goes:

[Hawk552]

Originally Posted by joaquimandrade How can increment a variable inside a forward be less or more expensive depending on which forward?

As far as I know, it's profiling how long it actually takes for it to execute the function from front to back. I don't think it includes how long it takes to actually find it, though.

Originally Posted by joaquimandrade What is the point of profiling forwards?

Well, we're trying to see whether FakeMeta's one forwards are faster than Engine's multi forwards. I think it's a good idea to benchmark these against each other.

[joaquimandrade]

Originally Posted by Hawk552 As far as I know, it's profiling how long it actually takes for it to execute the function from front to back. I don't think it includes how long it takes to actually find it, though. Well, we're trying to see whether FakeMeta's one forwards are faster than Engine's multi forwards. I think it's a good idea to benchmark these against each other.

What "a forward being faster than other" means?

[Hawk552]

Originally Posted by joaquimandrade What means "a forward being faster than other"?

If it takes less time for all of the equivalent functions to be found and then executed from front to back with identical code inside them, then you can consider that forward faster than the one you're comparing it to. Basically, what I'm saying is this:

As far as I know, AMXX keeps a "lookup table" of all public functions in each active plugin. I think it would be faster to go through this and look for "client_PreThink" in each plugin than it would be to navigate through the registered forwards in FakeMeta. I could be wrong, though; one reason being that multiforwards have to scan through each plugin and each function in them, which I would assume becomes more and more trivial as FM_PlayerPreThink gets registered more times. FM_PlayerPreThink, however, has to execute a separate forward for each plugin registering it.

I don't know the internal workings of AMXX's forward handling well enough to really say with total certainty which would be faster. What I've posted is just a guess. It's also hard to benchmark them against each other. As I said before, a better benchmark may be to check how long it takes between PreThink and PostThink in each module, assuming that each forward is executed in the same order each time. This would include how long it takes to look up the PostThink function. The profiler, however, isn't capable of checking times between functions and can only track how long it takes for a function to run to completion.

Basically, I think we'd need a developer to tell us. There's no real way of benchmarking them against each other properly.