[Fixed] [g3.5pre-44-g1455111dd] 50% performance regression since 3.3

Bugs that have been investigated and resolved somehow.

Moderator: Developers

Re: [g3.5pre-44-g1455111dd] 50% performance regression since

Postby Blzut3 » Sat Jun 16, 2018 9:31 pm

dpJudas wrote:blzut3: are you absolutely sure you built it in release mode? (not that I think you'd normally make this kind of mistake, but its worth asking just to be sure)

Yes. Debug mode cuts the frame rate in half again ~55fps for paletted and ~25fps for true color. The 3.3.2 and 3.4.x binaries that I'm using are on debian.drdteam.org.

Are you using the nvidia blob or nouveau on Debian (guessing blob since IIRC nvidia still hasn't released firmware to change the fan speed so GPU clock is forced low in nouveau)? It probably is some weird slow spot on Mesa, but of course there has to be something in GZDoom that changed since the old version ran faster on the same version of Mesa.
Blzut3
Pronounced: B-l-zut
 
 
 
Joined: 24 Nov 2004

Re: [g3.5pre-44-g1455111dd] 50% performance regression since

Postby Graf Zahl » Sun Jun 17, 2018 12:40 am

Blzut3 wrote:there has to be something in GZDoom that changed


Obviously. But since you are the only one in a position to examine, I only can ask you to do some tests to narrow it down.
User avatar
Graf Zahl
Lead GZDoom Developer
 
Joined: 19 Jul 2003
Location: Germany

Re: [g3.5pre-44-g1455111dd] 50% performance regression since

Postby Cacodemon345 » Sun Jun 17, 2018 2:11 am

GZDoom 3.4.1 is slow on my laptop using an nVidia GPU, running Windows, but GZDoom 3.3.2 works fine. Why it happens makes me think if it is a problem on GZDoom's side or on the laptop's side... :?
Cacodemon345
 
Joined: 22 Dec 2017
Discord: Cacodemon345#9151

Re: [g3.5pre-44-g1455111dd] 50% performance regression since

Postby Graf Zahl » Sun Jun 17, 2018 2:20 am

What's an "NVidia GPU"? And what's "slow"?
You have to give more info to analyze a problem.
Also, which values are different when looking at the output of "stat rendertimes"?
User avatar
Graf Zahl
Lead GZDoom Developer
 
Joined: 19 Jul 2003
Location: Germany

Re: [g3.5pre-44-g1455111dd] 50% performance regression since

Postby Cacodemon345 » Sun Jun 17, 2018 2:26 am

Graf Zahl wrote:What's an "NVidia GPU"? And what's "slow"?
You have to give more info to analyze a problem.
Also, which values are different when looking at the output of "stat rendertimes"?

I mean, it is running a NVidia GeForce 940MX and is using a Intel i5-8250U CPU. It slows down whenever a graphically-intensive action takes place in graphically-intensive mods, such as Project Brutality etc. In GZDoom 3.3.2 it was faster, but now it is slow.

My desktop computer, on the other hand, doesn't seem to be suffering from slowdown. (running AMD Radeon RX 460 and Intel i5-6500 with GZDoom 3.4.1)
Cacodemon345
 
Joined: 22 Dec 2017
Discord: Cacodemon345#9151

Re: [g3.5pre-44-g1455111dd] 50% performance regression since

Postby Phoebe Clankinstudge » Sun Jun 17, 2018 3:22 am

Hello!

Well, what I've noticed since 3.4.0 is an speed loss in the following case:

Imagine having a large square sector. In the middle of it, there are several smaller sectors. Now, if there are a number of dynamic lights in the large sectors, then fps rate drops down when compared to 3.3.2.

Now, if you break up the large square sector into smaller pieces, there is no drop down any more.

Kind regards
zhadoom
Phoebe Clankinstudge
 

Re: [g3.5pre-44-g1455111dd] 50% performance regression since

Postby Graf Zahl » Sun Jun 17, 2018 3:29 am

What graphics hardware does this happen on?
User avatar
Graf Zahl
Lead GZDoom Developer
 
Joined: 19 Jul 2003
Location: Germany

Re: [g3.5pre-44-g1455111dd] 50% performance regression since

Postby Phoebe Clankinstudge » Sun Jun 17, 2018 3:31 am

Hello!

gtx750
gtx950m
Phoebe Clankinstudge
 

Re: [g3.5pre-44-g1455111dd] 50% performance regression since

Postby Graf Zahl » Sun Jun 17, 2018 3:47 am

Do you have a test map for that effect?
User avatar
Graf Zahl
Lead GZDoom Developer
 
Joined: 19 Jul 2003
Location: Germany

Re: [g3.5pre-44-g1455111dd] 50% performance regression since

Postby Graf Zahl » Sun Jun 17, 2018 4:27 am

Graf Zahl wrote:Also, which values are different when looking at the output of "stat rendertimes"?



I just want to emphasize this. If you do not provide any numbers showing your problems I have no means investigating them.
So either provide screenshots with this info or post a log with the output of the 'bench' CCMD.
User avatar
Graf Zahl
Lead GZDoom Developer
 
Joined: 19 Jul 2003
Location: Germany

Re: [g3.5pre-44-g1455111dd] 50% performance regression since

Postby Cacodemon345 » Sun Jun 17, 2018 4:52 am

I may do that later. I'm at work at this moment, so I don't have my laptop with me.
Cacodemon345
 
Joined: 22 Dec 2017
Discord: Cacodemon345#9151

Re: [g3.5pre-44-g1455111dd] 50% performance regression since

Postby dpJudas » Sun Jun 17, 2018 5:16 am

Blzut3 wrote:Are you using the nvidia blob or nouveau on Debian (guessing blob since IIRC nvidia still hasn't released firmware to change the fan speed so GPU clock is forced low in nouveau)?

I'm using the nvidia blob. Using nouveau with a Nvidia card generally seems like a bad idea.
dpJudas
 
 
 
Joined: 28 May 2016

Re: [g3.5pre-44-g1455111dd] 50% performance regression since

Postby JJP » Sun Jun 17, 2018 4:53 pm

I observed a similar performance regression with the software renderer on my Debian Stretch (64-bit) system from GZDoom 3.3.2 to 3.4.1.

The CPU is an AMD Phenom II X4 945 (yeah, it's kinda old), and the GPU is an Nvidia GeForce GT 710. The graphics driver being used is Nvidia's proprietary blob, version 390.48. The screen resolution is 1366x768 in windowed mode.

The area I tested is the beginning of Doom II's map01. Here are the FPS values I observed with the various software rendering modes in both versions; first value is from 3.3.2 (with "Software Canvas" set to OpenGL), and the second value is from 3.4.1:
  • Paletted, non-softpoly: ~230 fps -> ~120 fps
  • Paletted, softpoly: ~110 fps -> ~60 fps
  • True-color, non-softpoly: ~140 fps -> ~55 fps
  • True-color, softpoly: ~50 fps -> ~52 fps
Unfortunately, I'm unable to get decent data from either 'stat rendertimes' or the 'bench' console commands because the numbers that are reported, besides the FPS and map coordinates in the case of the 'bench' command, are all 0 or 0.000 or similar, which is certainly wrong. I have no idea why that is the case. This occurs with both GZDoom versions tested. I do get proper data if the OpenGL renderer is active, but this issue is about the software renderer, and I don't get sane data with the software renderer in all cases.

I do compile from source, and I compiled Release builds for both 3.3.2 and 3.4.1. If any more information is needed, let me know.
JJP
 
Joined: 11 Aug 2005

Re: [g3.5pre-44-g1455111dd] 50% performance regression since

Postby Blzut3 » Sun Jun 17, 2018 5:27 pm

Regression also occurs on Mac OS X. Late 2009 Mac Mini (Core 2 Duo P7550 with nvidia geforce 9400) running OS X 10.11.6. Using the official 3.3.2 and 3.4.1 builds I go from 79fps to 49fps.

A few other quirks on the Mac Mini is that the screen wipe animation is very slow on 3.4.1 and when switching from OpenGL to software the menu fonts will still be mostly linear filtered (with some characters randomly being unfiltered). Same vice versa. Restarting fixes the texture filtering, but not the wipe speed.
Blzut3
Pronounced: B-l-zut
 
 
 
Joined: 24 Nov 2004

Re: [g3.5pre-44-g1455111dd] 50% performance regression since

Postby Graf Zahl » Sun Jun 17, 2018 5:36 pm

That actually sounds like a buggy driver with the fonts. It means that the change of texture sampler settings does not get picked up by the texture.
This hardware is too old, though, to bother with it.

I wouldn't be surprised if the buffer mapping is in some way responsible. That part seems to be so incredibly spotty on some systems that even looking at it strangely might make it throw up.
Do you think you can pinpoint the problem?
User avatar
Graf Zahl
Lead GZDoom Developer
 
Joined: 19 Jul 2003
Location: Germany

PreviousNext

Return to Closed Bugs

Who is online

Users browsing this forum: Google [Bot] and 1 guest