Polymer slowdown in newer versions
#1 Posted 06 July 2015 - 04:12 PM
EDIT: I tested with 2 more custom projects, Graveyard and Project Blue Skies, and they both showed a drop of about 3 FPS with the newer versions. Not nearly as bad as the double digit drops in WGR2 but a drop none-the-less. Hopefully someone else can verify.
This post has been edited by Mark.: 06 July 2015 - 04:58 PM
#2 Posted 06 July 2015 - 07:24 PM
r4773 | hendricks266 | 2014-11-28 18:29:42 -0800 (Fri, 28 Nov 2014) | 1 line
Fix POLYMER=0 GTK C++ build. DONT_BUILD.
r4701 | terminx | 2014-10-29 10:06:20 -0700 (Wed, 29 Oct 2014) | 1 line
Fix MSVC warning in Polymer
----
Possibly it isn't even Polymer related. Maybe one of the revisions adds extra processing, and because Polymer is so CPU intensive, that causes a framerate drop.
#4 Posted 07 July 2015 - 05:36 AM
Tea Monster, on 07 July 2015 - 02:10 AM, said:
And how on Earth did it take over 6 months before anyone noticed and reported it?
#5 Posted 07 July 2015 - 06:29 AM
#6 Posted 07 July 2015 - 09:53 AM
This post has been edited by Mark.: 07 July 2015 - 09:54 AM
#7 Posted 08 July 2015 - 02:35 AM
*Polygon Aversion Disorder
#8 Posted 15 July 2015 - 05:11 PM
This post has been edited by Sgt. Rarity: 15 July 2015 - 08:17 PM
#9 Posted 16 July 2015 - 05:01 AM
#10 Posted 18 July 2015 - 04:42 PM
This post has been edited by SamusBailey: 18 July 2015 - 04:43 PM
#11 Posted 18 July 2015 - 05:23 PM
#12 Posted 18 July 2015 - 05:54 PM
Trooper Dan, on 18 July 2015 - 05:23 PM, said:
80 revisions is a wide gap. Could you or Mark describe a specific procedure to get to whatever area renders measurably slower?
This post has been edited by Hendricks266: 18 July 2015 - 05:55 PM
#13 Posted 18 July 2015 - 06:35 PM
I've uploaded a few screenshots to show some visual glitches and FPS jumps which are huge in such a small area with little draw distance.
lowfps.jpg shows my frame rate sucks. Move a bit to the right and i get great frame rates in highfps.jpg
Stairs.jpg shows the location in my map where you can go upstairs and downstairs. However, for whatever reason it works in Eduke thank goodness but not in Mapster. Strange bug.
Bulletsholes invisible wall.jpg - Shows that for some reason Eduke isn't drawing the sector properly and is getting confused with some over lapping sectors. These are all taken from 20150716-5303 in Windows 7 64Bit using the 64bit Eduke synthesis. These are also problems in previous synthesis builds as well. All screenshots are from therock.map. I find this a good map to test performance issues as a result of the heavy usage of SOS and TROR.
As for generic game hitches I'm not sure i can comment on that as that hasn't been an obvious problem for me.
I'm not trying to hijack this thread but only contribute to performance related issues as well as some other issues which if cleared up might help the overall game performance.
This post has been edited by Paul B: 19 July 2015 - 05:41 AM
#14 Posted 19 July 2015 - 05:21 AM
This post has been edited by Paul B: 19 July 2015 - 05:34 AM
#15 Posted 19 July 2015 - 06:09 AM
Hendricks266, on 18 July 2015 - 05:54 PM, said:
In my case you would have to download and run my whole custom project and the first map.
Version 4593 at map start..... Polymer off = 175 FPS Polymer on = 67 FPS
Version 4777 at map start .....Polymer off = 175 FPS Polymer on = 20 FPS
In this case it is a relatively simple outdoor area that is isolated from the rest of the map by white walls and a closed ceiling door. Just 4 SE49 lights.
EDIT: and IIRC areas of poor performance in the large map drop from about 20 FPS in 4593 down to 13 FPS with 4477. So you can see why I need every drop of speed I can find. Luckily this is more of an adventure game and the low framerate in some areas won't get the player killed like if they were going up against a horde of aliens.
This post has been edited by Mark.: 19 July 2015 - 07:05 AM
#16 Posted 19 July 2015 - 12:12 PM
#17 Posted 19 July 2015 - 01:07 PM
#18 Posted 19 July 2015 - 03:32 PM
Tea Monster, on 19 July 2015 - 12:12 PM, said:
------------------------------------------------------------------------ r5266 | Plagman | 2015-06-14 18:42:31 -0500 (Sun, 14 Jun 2015) | 1 line add pr_nullrender variable to toggle drawing and updating GL buffers ------------------------------------------------------------------------ r5267 | Plagman | 2015-06-14 18:42:52 -0500 (Sun, 14 Jun 2015) | 1 line Fix automap lines in Polymer mode ------------------------------------------------------------------------ r5294 | Plagman | 2015-07-12 22:46:35 -0500 (Sun, 12 Jul 2015) | 1 line Polymer: store map data in a single buffer with a unified vertex type. ------------------------------------------------------------------------ r5295 | Plagman | 2015-07-13 11:01:18 -0500 (Mon, 13 Jul 2015) | 1 line Polymer: fix synthesis warning. ------------------------------------------------------------------------ r5296 | Plagman | 2015-07-14 02:08:44 -0500 (Tue, 14 Jul 2015) | 1 line Polymer: add preliminary support for batching draws. ------------------------------------------------------------------------ r5297 | Plagman | 2015-07-15 00:14:08 -0500 (Wed, 15 Jul 2015) | 1 line Polymer: fix regression with Y-flipping bit. ------------------------------------------------------------------------ r5298 | Plagman | 2015-07-15 03:23:00 -0500 (Wed, 15 Jul 2015) | 1 line Polymer: upload bucket indices through pinned host memory. ------------------------------------------------------------------------
#19 Posted 19 July 2015 - 04:01 PM
Tea Monster, on 19 July 2015 - 12:12 PM, said:
Maybe they're following the same policy as the MAME team: "Meh, who cares about our software getting slower and slower? Hardware's gettng faster and faster, that'll fix it!"
#20 Posted 19 July 2015 - 04:31 PM
Hendricks266, on 18 July 2015 - 05:54 PM, said:
I could not myself, but in my previous post I did note that there were only two revisions in that range (r4701 and r 4773) that mentioned polymer. Seems like that would be a good place to start.
#21 Posted 21 July 2015 - 04:49 PM
#22 Posted 21 July 2015 - 05:12 PM
Tea Monster, on 21 July 2015 - 04:49 PM, said:
They are optimizing all areas of code which overall will improve the overall fps since most of the game is heavily based on CPU. I've noticed gradual improvements over the years. It's just a slow progression and these dev's already have a lot on their plate. I'd like to know how you came up with 40fps loss that's pretty specific. I guess it would all depend on what features of Polymer are being used which contributes to the overall performance problems.
This post has been edited by Paul B: 21 July 2015 - 05:23 PM
#23 Posted 21 July 2015 - 06:00 PM
This post has been edited by Mark.: 21 July 2015 - 06:06 PM
#24 Posted 21 July 2015 - 06:48 PM
Mark., on 21 July 2015 - 06:00 PM, said:
Thanks Mark for the clarification. I accidentally skimmed over it.
This post has been edited by Paul B: 21 July 2015 - 06:49 PM
#25 Posted 21 July 2015 - 07:53 PM
Mark., on 21 July 2015 - 06:00 PM, said:
Maybe he can also address the issue with my maps on Polymost, where fps dropped from a descent rate to red numbers on later synthesis.
#26 Posted 21 July 2015 - 10:00 PM
#27 Posted 21 July 2015 - 10:20 PM
Altered Reality, on 19 July 2015 - 04:01 PM, said:
How obnoxiously obtuse. It's fine with me if you don't particularly understand software development and have no grasp of the concept that sometimes optimizations in certain areas equal regressions in others, but please, keep ignorant comments like that to yourself.
#28 Posted 21 July 2015 - 10:20 PM
Trooper Dan, on 21 July 2015 - 10:00 PM, said:
Is that really accurate? The vanilla maps wouldn't have the use of the new S.E for lighting \ smoke effects \ TROR which may take their toll on the FPS. I'm guessing where the slow downs may occur are based around these newer effects within the Polymer as oppose to the original level design. Since the original Duke maps weren't built upon Polymer's new features they probably won't exhibit the same behavior as mentioned in modern versions of the duke map file. Which may explain why Mark notices fps drops in his maps but not others.
For us non programmers (Speaking on behalf of myself) I can only speculate since I don't really know what's going on behind the scenes.
There was an instance for me when Paralax skies seemed to have a significant impact on FPS especially when the ceiling of the paralax texture meets the floor. But it could have just been an isolated incident with the last map I was working on as a result of heavy use of TROR and large open areas.
This post has been edited by Paul B: 21 July 2015 - 11:06 PM
#29 Posted 21 July 2015 - 10:54 PM
Paul B, on 21 July 2015 - 10:20 PM, said:
Sorry I wasn't being clear. What I meant by "vanilla" was maps that use the base tileset and base CON code. That could include user maps that take advantage of TROR and SE lights or even the HRP, so long as they don't include custom art/models or special code. My point was that we don't need to make it hard on the developers to isolate the problem. Any number of user maps that already work with the base game will do.
#30 Posted 22 July 2015 - 07:53 AM
TerminX, on 21 July 2015 - 10:20 PM, said:
That was a sober description of the actual philosophy followed by the MAME team, and they said it themselves LOTS of times in emulation forums, whenever someone complains about a performance decrease from one release to the next. The only part of my post that might have been inaccurate was assuming that the developers of the Polymer renderer also swear by that philosophy.