Profiling a plugin?

[Dave2001]

A while back Doom told me about this way of finding slow parts of code, called "profiling". I had never done this before. So, I went & checked "enable profiling", but the build->profile option was disabled. I assume that this is because it is a dll file. Is there any way to profile a plugin dll?

I don't believe I've never heard of profiling before...

[Hacktarux]

It's definitely something that i wanted to make when i have tried to improve the speed of the tr64 plugin. But i have never understand how it works with vc++.

And i remember that i have found something on the microsoft site that says it's a bug of vc++: on some system the profile option is disabled and they gave a .reg file to solve this problem but it didn't enable the option....

[Dave2001]

Hmm... I created a test exe & profiling seems to work just fine there. I created a test dll & the option was disabled again...

Maybe if I start the profiling from the pj64 source?

[Dave2001]

There seems to be ways to do it with dlls and such from command prompt, but the problem is that they don't create a .pbo file like they are supposed to...

[Dave2001]

Whoa!!! I somehow did it, but I'm not sure quite how yet... Here's some of the output (i didn't run anything when i used pj64)

I'll try to figure out what I did & post it here.

Func Func+Child Hit
Time % Time % Count Function
---------------------------------------------------------
0.000 0.0 0.000 0.0 1 CountCombine(void) (combine.obj)
0.000 0.0 0.000 0.0 1 INI_Close(void) (ini.obj)
0.000 0.0 0.000 0.0 1 INI_FindSection(char *) (ini.obj)
0.000 0.0 0.000 0.0 17 INI_GetInt(char *,int,int) (ini.obj)
0.000 0.0 0.000 0.0 1 INI_Open(void) (ini.obj)
0.000 0.0 0.000 0.0 1 _$E17 (main.obj)
0.000 0.0 0.000 0.0 1 _$E19 (main.obj)
0.000 0.0 0.000 0.0 1 _$E20 (main.obj)
0.000 0.0 0.000 0.0 2 _DllMain@12 (main.obj)
0.000 0.0 0.000 0.0 1 _GetDllInfo (main.obj)
0.000 0.0 0.000 0.0 1 _InitiateGFX (main.obj)
0.000 0.0 0.000 0.0 1 _ReadSettings (main.obj)
0.000 0.0 0.000 0.0 3 __mtlockinit (mtlock.obj)
...

[Doomulation]

Originally posted by Dave2001
Whoa!!! I somehow did it, but I'm not sure quite how yet... Here's some of the output (i didn't run anything when i used pj64)

I'll try to figure out what I did & post it here.

Func Func+Child Hit
Time % Time % Count Function
---------------------------------------------------------
0.000 0.0 0.000 0.0 1 CountCombine(void) (combine.obj)
0.000 0.0 0.000 0.0 1 INI_Close(void) (ini.obj)
0.000 0.0 0.000 0.0 1 INI_FindSection(char *) (ini.obj)
0.000 0.0 0.000 0.0 17 INI_GetInt(char *,int,int) (ini.obj)
0.000 0.0 0.000 0.0 1 INI_Open(void) (ini.obj)
0.000 0.0 0.000 0.0 1 _$E17 (main.obj)
0.000 0.0 0.000 0.0 1 _$E19 (main.obj)
0.000 0.0 0.000 0.0 1 _$E20 (main.obj)
0.000 0.0 0.000 0.0 2 _DllMain@12 (main.obj)
0.000 0.0 0.000 0.0 1 _GetDllInfo (main.obj)
0.000 0.0 0.000 0.0 1 _InitiateGFX (main.obj)
0.000 0.0 0.000 0.0 1 _ReadSettings (main.obj)
0.000 0.0 0.000 0.0 3 __mtlockinit (mtlock.obj)
...

That's function profiling.
It lists how many times your functions get called. It's useful to see if a function gets called lots, then you should try to optimize that function, and then you can see which functions that are NOT called. Those should be removed, saves you size

There's other profiling options as well, Dave. Try them, too.

Hacktarux: Profiling isn't hard at all to use. You just need to have an EXE to do it the easy way.

You simply goto options and select "enable profiling" in debug. Then goto build-->profile and you can select a method of profiling. When selected, click OK and the app will automaticlly compile and run. After it's done, you'll see the results of the profiling.

[Dave2001]

i can't get it to happen again

jeez... i wrote a simple hello world application & it still does not profile like they say it should...

PREP : fatal error PRF1011: cannot open file c:\programming\proftest\release\proftest .pbo

[Rice]

Dave, you can try Intel's Vtune 5.0 or 6.0, or if you can find, try TrueTime, they are professional profiler.

I don't think you can do multi-thread profiling in VC. I tried it before, it did not work, and I remember MS said VC only support single-thread profiling.

[Dave2001]

Thanks Rice, vtune works great.

25% of the time in Glide64.dll is spent calculating vertices (transformation, lighting), which is understandable

10% is in my CopyBlock function, which can be optimized lots

[Doomulation]

Originally posted by Dave2001
Thanks Rice, vtune works great.

25% of the time in Glide64.dll is spent calculating vertices (transformation, lighting), which is understandable

10% is in my CopyBlock function, which can be optimized lots

Yet, more speed increase!