Porting apitrace

@kas1e

Quote:
It matters because I need to understand what you're trying to achieve before I can give any advice.

CompileShader() is the function where shaders are passed to Warp3D Nova. However, remember that they are in SPIR-V format, and NOT the raw GLSL.

Hans

@Hans
Quote:

I need shaders in all their trasformation till the very latest sended directly to nova.

So far we can get those ones which we send to Ogles2 , and so we need those ones which ogles2 output to warp3dnova after glslnag-internal conversion happens with all the tweaks and workarounds putted in.

So yeah, as it will be SPIR-V format, then it's CompileShader(), thanks !

@Capehill
Is NOVA's filters works now ?

Why i ask, because i see in the code of ogles2_module.c that: GENERATE_FILTERED_PATCH

But in warp3dnova_module.c can see only those ones (as it was before filters were added): GENERATE_NOVA_PATCH

As well as when i test it previously, some output from w3dnova patches was still here, while i comment them all out.

@kas1e

Nova patches must be installed differently, that's why there is a separate macro for them.

I originally used strstr() for string matching (for convenience, but it turned out to be a bad idea). I changed the code to require an exact string match by using strcmp().

@thread

Probably overkill/not completely relevant but there was a recent Phoronix article on "RenderDoc" updates:

https://www.phoronix.com/scan.php?page ... px=RenderDoc-1.3-Released

Cheers!

glSnoop v0.1 is available: https://github.com/capehill/glsnoop/releases

@Capehill
I also asked Mason for some artwork, and he made glsnoop icon as well as drawer icon, so we can include them to the repo (if you doesn't mind of course).

There is:
http://kas1e.mikendezign.com/aos4/glsnoop/glsnoop_mason.zip

@kas1e

Nice! Will surely include them in the future. Thanks.

@Capehill
Tested latest commit48 of glsnoop, and found 2 issue (1 bug and one cosmetic issue):

bug:

Just run it as "glsnoop GUI" , and then run any gl4es app, it then crashes in warp3nova_dmodule at line 372, in W3DN_CreateFrameBuffer().

Ignore DSI help through, so it something easy should be (i hope).

cosmetic issue:

Also another error in GUI mode, is that counts of errors not realtime, i.e. you run something, exit, and error count show 0. Then you resize a little gui window, and only then error count refresh (to 9 in my case with prototype).

But gui window start to looks better anyway :) more informative.


ps. i am sure you consider it already, but don't forget about os4depot release :) glsnoop is matured enough for sure.

@kas1e

Thanks, I will investigate. EDIT: cannot reproduce DSI with Neverball, AmiCraft, Prototype, Foobillard+ or FrikinShark. Need more information...

Error counter refreshing is based on Intuition tick events which only active windows receive. Maybe timer.device should be used.

@Capehill
Maybe you on release kernel, and not on debug one ? But will post full crashlog a bit later.

As for error code refreshing: sure it should refreshes when window active, but i was active and didnt refreshes, refreshes only when i resize window a little

@kas1e

Got it. OGLES2 seems to pass nullptr for "errcode" parameter. Can be worked around. This also means that OGLES2 is not interested in error code for some reason?

@Capehill
Not sure as well, need to ask Daniel on ogles2 thread with test case maybe (?)

@Capehill
commit49 fix crash on startup , thanks !:



Btw, as i see it not "OGLES2 seems to pass nullptr for "errcode" parameter. Can be worked around. This also means that OGLES2 is not interested in error code for some reason?" , but NOVA instead, so probabaly that question/bug report to Hans instead.

@kas1e

It's OGLES2 who passes the nullptr to Nova, so this must be checked in Nova module before trying to dereference the pointer. It's anyway legal to pass a nullptr to Nova in these cases.

@Capehill
Tested latest versions, all works, and what is good profiling also works !

Through, what i note,is that if in gui mode it was paused/unpaused, then, on exit there is _NO_ profiling info (like pressing pause/trace break whole process of counting things).

For example for now i trying to understand why SuperTuxKart the same slow over gl4es as on minigl (i.e. no differences in most cases), and so, i run glsnoop in GUI mode, press "pause". Then, load game, start track, and then press "Trace", tracing a little for 5 seconds, and press "pause". Then exit from glsnoop, and in log didn't have profiling info, just usuall loggin.

But if i just run it in gui mode, and didn't press pause , and then just trace a little and exit, then profiling info is here.

@kas1e

Good, sounds like 0.2 release might be close.

Profiling info is only shown when application destroys the context, not when glSnoop quits. But this should be easy to change.

@Capehill
It can be and when app destroyed, and when glsnoop quit (if that make sense?)

But problem that after you hit pause, then profiling didnt saves. But if you didnt hit pause, then profiling saves even when you just quit from glsnoop without destroing application

@Capehill
Another idea which maybe you find worth of adding : adding something like profiler-only flag to command line , so, while all the functions will be patches and counted as usuall, log will not be writen anything, but only final profiling information.

By this way, we will be able to avoid slow-downs in heavy games when we want only to know, where all the loading happens, and how much of and of which ones calls are used.

Because when i only need profiling, and even if i redirect loggin via sashimy to RAM:, it still slowdown the games pretty much (not _THAT_ much as if it was writen to serial of course, but still, it reduce fps pretty much, and for some big-slow games, its hard to profile it as you want it).

@Capehill
Btw, in NOVA output i didn't see any call to nova's CompileShader(), while it for sure should be called, as OGLES2 one call glCompileShader() for sure.

For example that will be helpfull to see speed of executiong of the shaders when they passed to/from ogles2 , and how much time takes NOVA's CompileShader().

Probabaly simply adding CompileShader() to warp3dnova's filters list will not make it "just works" ?