Compileable version of Alien Breed 3D II - The Killing Grounds

[Karlos]

Originally Posted by DanyPPC

On MiST fpga (minimig AGA core):

I have a little question, excuse me if this is just talked, is this new TKG exe better than old TKGPatch ?
At least for a standard A1200 + Blizzard 1230/IV ?

I believe the current build is significantly more performant. Amipal did a comparison video he posted a few pages back.

[Karlos]

Originally Posted by grond

Yes, exactly. The V4SA beats the Pi even though it is slower at computing a frame because it copies the frame to superfast-chipmem in close to no time at all. The Pi has to interface a real chipmem bus and doesn't utilise the chipmem bandwidth to its max. The comparison to V500 and V1200 would be interesting because those are also slowed down by the real chipmem bus. Once the Pi can utilise the chipmem bus to its maximum throughput, it will certainly beat the Vampire accelerators. The V4 will probably still come out on top and could only be beaten by a Pi-standalone or WinUAE.

My take is that even if the max frame rate is slower, the FPS consistency over time for the Pi will be better. The C2P time is essentially constant, depending only on the screen area copied. The frame calculation varies significantly depending on scene complexity and that's where faster computation will win.

[Karlos]

Originally Posted by grond

Yes, exactly. The V4SA beats the Pi even though it is slower at computing a frame because it copies the frame to superfast-chipmem in close to no time at all. The Pi has to interface a real chipmem bus and doesn't utilise the chipmem bandwidth to its max. The comparison to V500 and V1200 would be interesting because those are also slowed down by the real chipmem bus. Once the Pi can utilise the chipmem bus to its maximum throughput, it will certainly beat the Vampire accelerators. The V4 will probably still come out on top and could only be beaten by a Pi-standalone or WinUAE.

My take is that even if the max frame rate is slower, the FPS consistency over time for the Pi will be better. The C2P time is essentially constant, depending only on the screen area copied. The frame calculation varies significantly depending on scene complexity and that's where faster computation will win.

Personally I'd rather have a rock solid 25fps than swings of 10 to 50.

[shelter]

Originally Posted by abu_the_monkey

@shelter

what speed is your TF1230 clocked at please?

50mhz/128MB

[Karlos]

@abu_the_monkey

Can you get a PR ready for this instrumentation?

I'm thinking to add a build flag soon so that you can build a fully instrumented version and a release version that doesn't include any of it for the best possible performance.

[DanyPPC]

Better to have 2 version, one for test purposes and one without data speed on screen for full performance.

Anyway I'm happy with this latest build. It has better performance on all my Amiga systems respect to original and TKGPatch executables.

Well done !

[shelter]

Originally Posted by Karlos

@abu_the_monkey

Can you get a PR ready for this instrumentation?

I'm thinking to add a build flag soon so that you can build a fully instrumented version and a release version that doesn't include any of it for the best possible performance.

Can't it be done with a togglekey in game?

[Karlos]

Originally Posted by shelter

Can't it be done with a togglekey in game?

It can. I suppose it comes down to wanting to be able to have a build that's as streamlined as possible. Also, there aren't many function keys left

[abu_the_monkey]

Originally Posted by Karlos

@abu_the_monkey

Can you get a PR ready for this instrumentation?

I'm thinking to add a build flag soon so that you can build a fully instrumented version and a release version that doesn't include any of it for the best possible performance.

I have now raised a PR but it probably needs a tidy up and a proper check from yourself or pipper as I may well not be doing things in the most efficient way.

[Weasel Fierce]

Just wanted to thank you guys again for all the amazing work.

[abu_the_monkey]

Originally Posted by Karlos

I'm thinking to add a build flag soon so that you can build a fully instrumented version and a release version that doesn't include any of it for the best possible performance.

sounds like a good idea to me

[Karlos]

Originally Posted by abu_the_monkey

sounds like a good idea to me

I want to add a bunch of flags for turning on/off the rendering of different object classes at runtime for the "dev" build, e.g. floors, flat shaded walls, goraud walls, doors, lifts, bitmap sprites, glow sprites, blend sprites, lightsource sprites.

Even without recording the time for each class, it can be deduced using the overall frame draw time now. Adding these features is definitely a reason to make it a compile time flag because I'll be adding a lot of indirection.

[abu_the_monkey]

Originally Posted by Karlos

I want to add a bunch of flags for turning on/off the rendering of different object classes at runtime for the "dev" build, e.g. floors, flat shaded walls, goraud walls, doors, lifts, bitmap sprites, glow sprites, blend sprites, lightsource sprites.

Even without recording the time for each class, it can be deduced using the overall frame draw time now. Adding these features is definitely a reason to make it a compile time flag because I'll be adding a lot of indirection.

yes I agree, having a whole bunch of extra stuff in there is not going to help with performance so a dev and release build seems a reasonable option.

[Karlos]

Originally Posted by abu_the_monkey

yes I agree, having a whole bunch of extra stuff in there is not going to help with performance so a dev and release build seems a reasonable option.

Also, I already started...

[abu_the_monkey]

[amipal]

Originally Posted by Karlos

I believe the current build is significantly more performant. Amipal did a comparison video he posted a few pages back.

Here we go:
[ Show youtube player ]

[AlphaAmiga]

Just ran a quick 5 min test on my Rev 6 060 (TF1260 128MB) clocked at about 94MHz.

This was done in full screen mode.

In the first room of the first level the FPS was around 19-20, dropping to 10-11 FPS when the first door opens, where it quickly wobbled back up to about 18 FPS initially outside the room.

It "settled" around 15-16 FPS when walking up the first corridor towards the big red blast door.

If I approach the door with the devil dog behind it, FPS again dips to about 10FPS and then climbing back up to about 15FPS in the corridor.

Didn't want to run the CPU for too long at that clock seed as although I have active cooling on the CPU, I have never run it that fast before and didn't want to take the risk, until I am able to ensure good air-flow in the case.

I do also have a Mediator & Voodoo 3 equipped A1200 060 tower system (Rev 1 060 on a TF1260 which I can clock to about 64MHz with 128MB Fast Ram - damn Rev 6 prices !). I can try whichever build of the game people think would be best on that too if anyone's interested?

[abu_the_monkey]

@AlphaAmiga
yes please,
could you try the last build from post #1400 and post a photo of it when paused ('p' when in the game) in the room you start level A in?
do you have a copy of sysspeed? it would be interesting to know what the memory on the TF1260 performs like.
the slowdown near doors (and lifts) is due to the engine rendering what is behind it and then rendering the door (or lift) over the top effectively rendering everything twice (it's a known issue) causing the frame rate sink like a rock.

[abu_the_monkey]

Originally Posted by Karlos

Do we have any real copy ram to VRAM numbers for RTG systems available. I recall my BVision being 14MB/s but I no longer recall if that is copy speed or just write speed.

there is a tool on aminet that may return this info. if not, it comes with the source code so maybe could be adapted for what you want.

http://aminet.net/search?query=cgxbenchmark

[AlphaAmiga]

I am currently running memory tests using sysspeed on both my systems at serval clock speeds.

I will also look at running the cgxbenchmark program, but whilst I am at this, are their any other benchmarks you guys would be interested in from sysspeed?