| |
| |
| |
|
Page: 1 2
Comments:
<0> oi
<1> 30 if im lucky
<0> seems hard to imagine ..
<0> well, your choice if you want to put the effort into optimizing that .. but i guess your idea to collect glvertex and issue a gldrawelements seems reasonably
<0> *reasonable
<1> not sure if i can do that tbh
<1> cause i cant effect the client states
<1> like client side vertex pointers etc
<0> hm?
<1> immediate mode is also a pain in the *** because
<0> you mean, you must make sure you dont change it?
<1> yeh, otherwise will mess up compatability
<1> also immediate mode is annoyingly flexible
<0> ya, but you can just keep track, cantyou?
<0> all calls go thru your wrapper anyway
<1> ie u can specifiy colours every vertice, or like every 3 vertices or whatever
<1> yeh ure right
<1> i could do perhaps
<0> true
<1> brb
<2> Drivers already buffer immediate calls and dispatch them in bundles
<2> Unless you have some very akward or obsolete drivers
<0> so hes still left with the per-call processing he is doing for glVertex et al .. maybe dukey should try to optimize their code first as tiny improvements should have quite an impact when called so often
<0> (their code = code of those wrapper functions)
<1> well
<1> im not doing anything to the vertices, im just supplying a new projection matrix for off axis projected view
<0> for example, depending on your compiler, you could set options so that no frame is set up
<0> less pushing and copying of registers .. maybe not a big deal for one call, but as you go into the thousands etc .. try it
<1> no frame?
<0> stack frame
<1> u meam inlining my functions?
<0> nono
<1> mean
<1> ?
<0> when your function is called, it stored the ebp register and sets it up, so that ebp + x points to local variables and ebp - y (note the plus and minus difference) points to p***ed arguments .. now it may as well be the other way round with those plus and minuses
<0> *stres
<0> *stores
<1> ok
<0> if you tell the compiler to not generate this code .. you gain something
<1> i am only familar with ARM ***embly
<1> that might help
<1> x86 processors do have more registers, but they are virtual
<1> programmer cant access them
<0> been some time since i did asm .. so i might have mixed up some details
<0> well, you can access eax-edx, ebp and two others i cant remember the name of
<1> is this a compiler option ?
<1> cause i cant write x86 ***embly :p
<0> oh nono, the compiler will use the registers as they fit
<1> well tahts proably with the problem with x86, its only got 8 :p
<0> it depends on the compiler of course, but i doubt you have that much options
<0> right but thats enough .. which compiler are you using?
<1> visual studio 2003
<1> i can show u some of my code ..
<1> hang on
<1> finds paste board
<0> have a look into naked functions (see msdn), not sure if they are of help tho ..
<0> and then theres also an option in the compiler that does the stack frame thing
<0> in the Optimization part in the c++ section
<0> "/Oy (Frame-Pointer Omission)"
<0> "This option speeds function calls, because no frame pointers need to be set up and removed. It also frees one more register, (EBP on the Intel 386 or later) for storing frequently used variables and sub-expressions."
<0> im just reading: full optimization implies that already, unless you explicitly disable it with /Oy-
<1> ok lets have a look
<0> also have a look thru the other compiler options .. maybe you can tweak something there
<1> omit frame pointers = no
<1> should i change that ?
<0> yip
<1> change to yes ?
<0> whats the optimization leven?
<0> *level
<0> yip
<1> i'll take screenshot
<0> okay
<1> can u accept dcc
<1> or shall i imageshack it :p
<0> nope sorry .. try uploading it
<1> k
<0> ya,do that .. firewall .. sorry
<1> hang on
<1> tis being slow
<0> no worries
<1> right
<1> http://img230.imageshack.us/img230/222/untitledsw7.png
<1> ^
<0> oi
<1> ?
<1> ;p
<0> wait, youre not working with the debug build are you?
<1> no :p
<0> dont change anything there
<1> ok
<1> opps i took screenshot of debug
<0> at the top left, you can switch to the release options
<1> yeh
<1> sec
<1> http://img230.imageshack.us/img230/4913/untitledur5.png
<0> well, try finding out whether "Maximize Speed" should be changed to "Full Optimization" .. i would guess so
<1> worth a try
<1> never really touched those options before
<0> then set "Omit Frame Pointers" to yes .. tho that should be implied, try anyway
<1> right
<1> lets try out release build speed
<1> hm weird
<1> favour side or speed
<1> did speed
<1> came out much smaller :S
<0> *g*
<0> try also global optim. to yes
<1> hm weird
<0> ok, according to msdn, setting "optimization" to "maximize speed" instead of "full optimization" should be better
<1> not sure that didn't make it slower
<0> but im looking at the msdn for vs05
<0> so maybe set it back to maximize speed and leave "omit frame pointers" at yes
<0> hm, we're pretty much at the beginning already .. if you used /O2 already, you should have /Oy also enabled already: "imply the use of the Frame-Pointer Omission" is what it says for /O2
<1> hang on
<0> brb .. dinner time
<1> hm bit better
<1> brb
Return to
#opengl
or
Go to some related
logs:
#microsoft
#computers
#nhl
#unix
conexant amc audio window vista
scansdisk
WEBSITE CLOCKS
#gamedev
#politics
only 128 meg system memory
|
|
