my god. if only my old 486 had that -potato option in 1994... the 8fps I was suffering with could've been a nice 25 😳

Wow, the timing could not be better. I tried regular Doom for the first time last night on my 386DX/40 build, just for a quick test and a laugh. Will test and report back. Thanks!

Pierre32 wrote on 2020-06-24, 00:21:

Wow, the timing could not be better. I tried regular Doom for the first time last night on my 386DX/40 build, just for a quick test and a laugh. Will test and report back. Thanks!

I would also like to hear. I just dusted off my 386dx40. It has 8meg ram at the moment. Thanks

So assuming you run this on normal details how much faster is it from stock DOOM?
This whole project looks to be very nice, kudos!

I would love to hear all the details and tradeoffs about all the new optimizations.

This is truly awesome. Although I will never play it, because I have no need, I have utmost admiration for this kind of quality work and dedication. 😀

1901 realtics vs. 2193 realtics on my 486 [Asus VL/I-486SV2GX4, DX2 @ 66MHz, 1MB L2 Cache, 24MB RAM, 1GB CF HDD, Cirrus Logic 5428 VLB, ARGUS Prototype Rev. 02 #0], settings are the "usual Doombench" ones (Screen scaled down to status bar + 1 notch border, high detail, -nomusic -nomouse -timedemo demo3)
Timing is perfect here as well ... recently considered swapping the VLB Cirrus Logic graphics card against a Mach32, but was struggling with a decision when I saw the pricetag for those cards.

Thanks for this.

xcomcmdr wrote on 2020-06-24, 09:04:

I would love to hear all the details and tradeoffs about all the new optimizations.

No F11 gamma correction seems to be one i've noticed.

Interesting! And FPS ingame viewer is amazing thing!

keropi wrote on 2020-06-24, 04:33:

So assuming you run this on normal details how much faster is it from stock DOOM?
This whole project looks to be very nice, kudos!

In my testings using high detail + full screen + status bar and audio enabled it gains around 15% performance on a 486DX-50 with a VLB video card (Cirrus Logic). Including other settings such as flat visplanes or limit distance view for objects increments even more that performance. Extracting performance it's really hard as the original code was already well optimized.

leileilol wrote on 2020-06-24, 13:17:

xcomcmdr wrote on 2020-06-24, 09:04:

I would love to hear all the details and tradeoffs about all the new optimizations.

No F11 gamma correction seems to be one i've noticed.

Oopps! Forgot to mention that one. It's a little bit faster without gamma correction, and I always felt that gamma correction made the game look really bad. The idea is to keep the port compatibility as high as possible and remove only barely used features. I will add wiki pages in github explaining all the options in detail with images, and how much performance you gain by using them.

MAZter wrote on 2020-06-24, 17:30:

Interesting! And FPS ingame viewer is amazing thing!

Kudos to @Jnechaevsky, I saw her implementation in Russian Doom and ported it to FastDoom, but using fixed point math instead of floating point math. I will make it show more stats in the future (memory usage, column and spans render calls, ...)

shock__ wrote on 2020-06-24, 11:38:

1901 realtics vs. 2193 realtics on my 486 [Asus VL/I-486SV2GX4, DX2 @ 66MHz, 1MB L2 Cache, 24MB RAM, 1GB CF HDD, Cirrus Logic 5428 VLB, ARGUS Prototype Rev. 02 #0], settings are the "usual Doombench" ones (Screen scaled down to status bar + 1 notch border, high detail, -nomusic -nomouse -timedemo demo3)
Timing is perfect here as well ... recently considered swapping the VLB Cirrus Logic graphics card against a Mach32, but was struggling with a decision when I saw the pricetag for those cards.

Thanks for this.

If you disable also the sound, the performance goes from 15% to 25% as the original sound code does lot's of weird things. Mainly it dumps all cached sounds that are not used after ~0.4 seconds, and does this check in every tick for all the channels. This slows down the performance and i'm looking to solve it.

Huge thanks to everyone, I wasn't expecting so many comments! Any ideas or criticism will be welcome 😄. One question, how many of you do use the mouse to play Doom?

Yes, mouse for sure. I really love that Y axis is disabled here, so I don't have to run Novert separately.

(Busy week, so I still haven't actually had time to test yet.)

Never used the mouse at all back in the day.
But if WASD is possible, then yes.

Very cool! I recently finished a full Ultra-Violence run of Doom 1 on my 486 DX4 120MHz/VLB and it really struggled with some maps in Episode 4 (Thy Flesh Consumed), so I'll be giving that a try with your port.

The removal of 'Y-axis mousewalk' is a big improvement in any case. I had to turn mouselook off in some sections where you had to walk on narrow ledges to reach some secrets.

[edit]
It runs very well. 3612 realtics (FastDoom with DOS/32A Extender v9.1.2) vs 3708 (FastDoom with DOS4GW v1.97) vs 4178 (Ultimate Doom v1.9) on Ultimate Doom's demo3.

And E4M2, which ran so bad I had to reduce the screen size on my UV run, is actually playable now 😀

Nice! Gotta try it out!
Arrow keys are for 1000 IQ gamers!

This is also really interesting when you try to run DooM in a web browser. All optimizations help a lot when it runs in WASM.

Playing with the shareware 1.9 IWAD does not display the third episode's title for some reason on the episode selection screen.

Is Doom faster when it uses a chained hash table instead of linear search?

I confirm that on the laptop 486 DX2 50Mhz fps increased by 1.52 (from 17.62 to 19.20) using Dos Benchmark Pack config for faster PC.

The video/source files makes it look like you're using the DMX sound library?