Reply 720 of 862, by Maraakate
- Rank
- Oldbie
Q2 Code has been around since about 2001 or so I think.
I haven't added any model lerping or any frills. But yes, I did compile with -O0. Good catch.
Q2 Code has been around since about 2001 or so I think.
I haven't added any model lerping or any frills. But yes, I did compile with -O0. Good catch.
Edit: I edited post to show the difference in speed is ~20% (probably nearer 15%) between glquake and qdos/fx. I had an anomalous reading before. 😀 Doesn't seem too far off from the difference between q2dos/fx and the ICD by 3dfx.
I believe it's faster than the software renderer at resolution along with the added bonus features.
Recompiled with -O2 and all that good stuff. Bigass1.dem gets was 67.0fps at 640x480 on the p3 550 with the voodoo 3.
The only other thing is verifying the cvars, but I couldn't find anything obvious that would impact performance and that isn't in qlquake. Not that I'd know with any certainty, but it's an impression.
gl_subdivide_size makes a difference at the expense of the sky looking slightly odd. Setting it to 1024 looked ok in the start map, I didn't try other places though.
Most of the cvars require a map reload.
I'll take a look at adding back in ARB multitexturing. Maybe in this game it might make a positive difference, you never know...
Thanks for the information. I think the result seems good, but as you said, the renderer can be a bit less efficient between the engines.
I'm also playing with gl_flashblend 0. This might slow things down a bit too. with it at it's default it makes those big ugly halos for rocket explosions and shit. I think it's intended to use that simple effect instead of the dynamic lights for a speed boost.
Ok, good. I'll check back after recharging the batteries, but thanks for bringing us the gl version of qdos! It looks great.
Edit: I'll try the flashbend variable, too.
That made another 10% difference.
What are you testing this on?
I know it's not accurate, just using dosbox/V1 versus glquake in dosbox/95/V1. I've benchmarked the latter many times so I have an expectation. But it will likely not transfer to performance on real machines.
Edit: -bpp 16 must be default?
Would the 3Dfx Global Palette extension have any impact? I noted it is mentioned in the readme for mesa and glquake. Edit: README.3DFX.
Edit2: on further testing, probably it's closer to 30% than 20% for performance difference.
3dfx global palette hard locks in DOS so it went away.
I did just port fitzquake to dos, 🤣. I'll test out the performance differences.
It's actually slower, but it does look a lot nicer. Stock Quake 1 is playable on it.
Here is the binary if you are interested.
Thank you! I look forward to trying it (especially on real hardware).
Edit: I tried it and it errors on missing sst hardware. Perhaps it is looking for a different 3dfx card or the emulation in dosbox is not complete. I'm looking forward to seeing the image quality though.
Edit2: from the binary's strings, I would guess it's looking for a voodoo3/4/5?
Oh, I'm sorry I compiled a voodoo 3/4/5 edition.
Also, you were right about the -O2. Now I can finally play warpspasm at acceptable framerates. I also tested against windows 98 with the opengl ICD and speeds are the same as in DOS, which is a good thing!
I posted the source, https://bitbucket.org/maraakate/fitzquake-dos/ , have fun. Now, there is a lot of warnings about implicit declarations, but the linker will find them all. I plan to fix that at some point.