Click Here
home features news forums classifieds faqs links search
6071 members 
Amiga Q&A /  Free for All /  Emulation /  Gaming / (Latest Posts)
Login

Nickname

Password

Lost Password?

Don't have an account yet?
Register now!

Support Amigaworld.net
Your support is needed and is appreciated as Amigaworld.net is primarily dependent upon the support of its users.
Donate

Menu
Main sections
» Home
» Features
» News
» Forums
» Classifieds
» Links
» Downloads
Extras
» OS4 Zone
» IRC Network
» AmigaWorld Radio
» Newsfeed
» Top Members
» Amiga Dealers
Information
» About Us
» FAQs
» Advertise
» Polls
» Terms of Service
» Search

IRC Channel
Server: irc.amigaworld.net
Ports: 1024,5555, 6665-6669
SSL port: 6697
Channel: #Amigaworld
Channel Policy and Guidelines

Who's Online
12 crawler(s) on-line.
 126 guest(s) on-line.
 1 member(s) on-line.


 A1200

You are an anonymous user.
Register Now!
 A1200:  4 mins ago
 OlafS25:  18 mins ago
 outlawal2:  38 mins ago
 AndreasM:  41 mins ago
 sibbi:  50 mins ago
 saimo:  1 hr 4 mins ago
 DiscreetFX:  1 hr 5 mins ago
 vox:  1 hr 16 mins ago
 kolla:  1 hr 27 mins ago
 Mr_DBUG:  1 hr 28 mins ago

/  Forum Index
   /  Amiga Development
      /  Packed Versus Planar: FIGHT
Register To Post

Goto page ( Previous Page 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 | 15 | 16 | 17 | 18 | 19 | 20 | 21 | 22 | 23 | 24 | 25 | 26 | 27 | 28 | 29 | 30 | 31 | 32 | 33 | 34 | 35 | 36 | 37 | 38 | 39 | 40 | 41 | 42 Next Page )
PosterThread
Gunnar 
Re: Packed Versus Planar: FIGHT
Posted on 5-Oct-2022 12:51:48
#421 ]
Regular Member
Joined: 25-Sep-2022
Posts: 471
From: Unknown

@Karlos

Quote:

I wasn't talking about PPC Amiga in the first place. I was talking about PPC in general.


Does OS4 run on Macbooks meanwhile?

 Status: Offline
Profile     Report this post  
Karlos 
Re: Packed Versus Planar: FIGHT
Posted on 5-Oct-2022 13:00:17
#422 ]
Elite Member
Joined: 24-Aug-2003
Posts: 4394
From: As-sassin-aaate! As-sassin-aaate! Ooh! We forgot the ammunition!

@Gunnar

Nobody but you is talking about OS4, but to the best of my knowledge, no, it does not. Are there any other irrelevant deflections you want to raise?

I'll raise one. I don't think you are actually Gunnar. He'd have provided the results of the exact test case I described earlier, even if it were only for the 68080 platform, pages ago. He wouldn't resort to throwing around irrelevant indirection because as an engineer he'd understand exactly what was being asked for, even the unstated parts, and deliver, to defend his claim.

You sir, are an imposter.

Last edited by Karlos on 05-Oct-2022 at 01:07 PM.

_________________
Doing stupid things for fun...

 Status: Offline
Profile     Report this post  
Gunnar 
Re: Packed Versus Planar: FIGHT
Posted on 5-Oct-2022 13:13:51
#423 ]
Regular Member
Joined: 25-Sep-2022
Posts: 471
From: Unknown

@Karlos
Quote:

Nobody but you is talking about OS4, but to the best of my knowledge, no, it does not

So you are talk about running the Mac with MorphOS, aren't you?

Last edited by Gunnar on 05-Oct-2022 at 01:20 PM.

 Status: Offline
Profile     Report this post  
Gunnar 
Re: Packed Versus Planar: FIGHT
Posted on 5-Oct-2022 13:19:18
#424 ]
Regular Member
Joined: 25-Sep-2022
Posts: 471
From: Unknown

@Cool_amigaN

Hello Amigan!

Quote:

the MorphOS G4@1.6Ghz


What hardware system is this that you use?

 Status: Offline
Profile     Report this post  
Karlos 
Re: Packed Versus Planar: FIGHT
Posted on 5-Oct-2022 13:22:53
#425 ]
Elite Member
Joined: 24-Aug-2003
Posts: 4394
From: As-sassin-aaate! As-sassin-aaate! Ooh! We forgot the ammunition!

@Gunnar

Quote:

Gunnar wrote:
@Karlos
Quote:

Nobody but you is talking about OS4, but to the best of my knowledge, no, it does not

So you are talk about running the Mac with MorphOS, aren't you?



I don't care which operating system as long as it runs. You can use OSX, YDL or whatever. If we're dealing with a command line application written in portable C with some possible ASM mixed in, it really doesn't matter.

_________________
Doing stupid things for fun...

 Status: Offline
Profile     Report this post  
Gunnar 
Re: Packed Versus Planar: FIGHT
Posted on 5-Oct-2022 13:34:05
#426 ]
Regular Member
Joined: 25-Sep-2022
Posts: 471
From: Unknown

@Karlos

Quote:

I don't care which operating system as long as it runs. You can use OSX, YDL or whatever. If we're dealing with a command line application written in portable C with some possible ASM mixed in, it really doesn't matter.


OK, I see you think the same CPU will always score the same?
Let me explain to you why the OS will make a difference.

The PowerPC memory controller has changeable settings. These setting will affect its behavior and will affect its performance. Also the PowerPC MMU mapping will influence the performance of memory operation. This means that both the used Bios and used OS do both have an impact on the performance. This means if you benchmark then the used OS does matter.

Last edited by Gunnar on 05-Oct-2022 at 01:44 PM.

 Status: Offline
Profile     Report this post  
Karlos 
Re: Packed Versus Planar: FIGHT
Posted on 5-Oct-2022 13:42:57
#427 ]
Elite Member
Joined: 24-Aug-2003
Posts: 4394
From: As-sassin-aaate! As-sassin-aaate! Ooh! We forgot the ammunition!

@Gunnar

This shouldn't matter if the operating system is intended for the hardware, you can assume it's set correctly. If it's not, that's an issue outside our control, but if it's a simple, portable (excluding the fact it's PPC+AltiVec dependent) program it can be recompiled and tested on different configurations.

_________________
Doing stupid things for fun...

 Status: Offline
Profile     Report this post  
Gunnar 
Re: Packed Versus Planar: FIGHT
Posted on 5-Oct-2022 14:35:41
#428 ]
Regular Member
Joined: 25-Sep-2022
Posts: 471
From: Unknown

@Karlos

If you want to get a feeling of your systems memory performance than you can use BUSTEST from Aminet - if you use an AMIGA OS.

If you run Linux on PowerPC then I can give you the following bechmark:
http://apollo-core.com/downloads/stream_ppc.exe

This is a PowerPC Linux memory benchmark.
The test will measure READ, WRITE, CMP, and COPY performance.
The test includes many different versions of each test, including versions using ALTIVEC instructions.
There are also tuned routines with optimizations for certain CPU model as (Efika)G2/G3/G4/G5 and CELL.


Happy Benchmarking!

Last edited by Gunnar on 05-Oct-2022 at 02:50 PM.

 Status: Offline
Profile     Report this post  
Karlos 
Re: Packed Versus Planar: FIGHT
Posted on 5-Oct-2022 14:57:11
#429 ]
Elite Member
Joined: 24-Aug-2003
Posts: 4394
From: As-sassin-aaate! As-sassin-aaate! Ooh! We forgot the ammunition!

@Gunnar

Forget the PPC version since you are so worried about the technicalities of it. Focus only on implementing the 68080/AMMX tuned version.

To reiterate:

Using the fastest available memory (e.g. does not have to be framebuffer) with whatever the most appropriate alignment for your system, blend a full size 1080p 32-bit ARGB SOURCE buffer onto an equal sized 32-bit ARGB DEST buffer such that EVERY pixel from SOURCE is blended with the corresponding pixel from DEST, storing the result in DEST. Use the blending equation DST_RGB = (SRC_ALPHA * SRC_RGB) + ((1 - SRC_ALPHA) * DST_RGB). DST_ALPHA does not have to be preserved or set to any particular value.

Record the time taken in milliseconds, to a precision of at least 1 millisecond, perhaps using the best of N sequential runs to make sure everything is hot.

Is any part of this unclear?

_________________
Doing stupid things for fun...

 Status: Offline
Profile     Report this post  
Gunnar 
Re: Packed Versus Planar: FIGHT
Posted on 5-Oct-2022 15:08:17
#430 ]
Regular Member
Joined: 25-Sep-2022
Posts: 471
From: Unknown

@Karlos

You might have missed it, but for 68K all you need is 4 instructions. I posted you the code before!

We are now waiting for your code

 Status: Offline
Profile     Report this post  
Gunnar 
Re: Packed Versus Planar: FIGHT
Posted on 5-Oct-2022 15:12:39
#431 ]
Regular Member
Joined: 25-Sep-2022
Posts: 471
From: Unknown

@Karlos

Quote:

Record the time taken in milliseconds, to a precision of at least 1 millisecond


Taking time on 68080 is very easy.
The CPU has in internal Clock register that you can read out with a simple MOVEC
You can measure time very precise this way with 10ns granularity.

 Status: Offline
Profile     Report this post  
Karlos 
Re: Packed Versus Planar: FIGHT
Posted on 5-Oct-2022 15:28:11
#432 ]
Elite Member
Joined: 24-Aug-2003
Posts: 4394
From: As-sassin-aaate! As-sassin-aaate! Ooh! We forgot the ammunition!

@Gunnar

You mean this?
Quote:

.loop
load (a0),D0
mulalpha (a1)+,D0
store D0,(a0)+
dbra.l d1,.loop


I don't know about AMMX mnemonics but this looks a bit suspect to me. My understanding was that AMMX, like MMX, reuses the FPU registers for performing vector operations. Unless your register names are case sensitive and D0-D7 are used to refer to them. That just seems a recipe for unnecessary confusion. It's also convenient that none of the operations specify the operand size and you just happen to have a mulalpha menomic that operates on ARGB. Lucky I didn't ask for RGBA, I guess.

But I digress. Run it then, chop chop.

Last edited by Karlos on 05-Oct-2022 at 03:43 PM.

_________________
Doing stupid things for fun...

 Status: Offline
Profile     Report this post  
Karlos 
Re: Packed Versus Planar: FIGHT
Posted on 5-Oct-2022 15:32:25
#433 ]
Elite Member
Joined: 24-Aug-2003
Posts: 4394
From: As-sassin-aaate! As-sassin-aaate! Ooh! We forgot the ammunition!

@Gunnar

Quote:

Gunnar wrote:
@Karlos
We are now waiting for your code


Don't be a moron. I have already explained. I don't have an altivec machine. Nevertheless the requirements of the test are hopefully simple enough that someone here who does will get to it. The real Gunnar would have done it by now, btw.

And let us not forget, that even if I did have a working Altivec machine, it's your claim that the 68080 is faster, so the burden of proof rests with you. There's no weaselling out of that, I'm afraid.

Last edited by Karlos on 05-Oct-2022 at 04:42 PM.
Last edited by Karlos on 05-Oct-2022 at 04:42 PM.
Last edited by Karlos on 05-Oct-2022 at 04:41 PM.

_________________
Doing stupid things for fun...

 Status: Offline
Profile     Report this post  
Gunnar 
Re: Packed Versus Planar: FIGHT
Posted on 5-Oct-2022 17:08:23
#434 ]
Regular Member
Joined: 25-Sep-2022
Posts: 471
From: Unknown

@Karlos

Quote:

Don't be a moron


Many thanks for being so friendly.

Quote:

I have already explained. I don't have an altivec machine. Nevertheless the requirements of the test are hopefully simple enough that someone here who does will get to it. The real Gunnar would have done it by now, btw.


Did you not propose that we both each do a benchmark?
To proof whether Vampire has faster memory interface or AmigaOne XE?

Of course we already know the outcorme we already have the results of the memory performance.
As we have done the memory benchmarks on these system before: stream, minibench and BUSTEST. BUSTEST scores 500-600 MB/sec on the Vampire.
What does it again score on AmigaOne, please remind me was it 100 MB?


Now you even want me to write your benchmark. You are kidding.
And when I do this then you claim that I cheated and coded the PPC code slow on purpose.
I think I know your style now...

Before this lowers down in just a waste of time, how about we make this a little more interesting?

How about we bet some real money on this?
How about we bet $200 or maybe more if you like?
What do you think?



Actually the real Gunnar is now preparing for A37.
A37 is the biggest Amiga event.

Of course the Apollo Team has a major booth there with 10 tables,
with over hundred of new Accelerators,
we will be giving coding lesson there
and we will present some awesome new Amiga games there.



Will anyone of you guys be at A37 Amiga meeting ?
and maybe even bring his AmigaOne with him?
We can code and run the benchmark there together. How does this sound?
And with more eyes we can verify the results together.
And have a beer together.

What do you think?

Last edited by Gunnar on 05-Oct-2022 at 05:26 PM.
Last edited by Gunnar on 05-Oct-2022 at 05:24 PM.
Last edited by Gunnar on 05-Oct-2022 at 05:17 PM.

 Status: Offline
Profile     Report this post  
Karlos 
Re: Packed Versus Planar: FIGHT
Posted on 5-Oct-2022 19:43:26
#435 ]
Elite Member
Joined: 24-Aug-2003
Posts: 4394
From: As-sassin-aaate! As-sassin-aaate! Ooh! We forgot the ammunition!

@Gunnar

Quote:

Gunnar wrote:
@Karlos

Quote:

Don't be a moron


Many thanks for being so friendly.


Apologies.

Quote:
Did you not propose that we both each do a benchmark?
To proof whether Vampire has faster memory interface or AmigaOne XE?


No I did not propose that at all, as anyone else reading the thread can attest. I proposed that you proved your original claim.

Nor did I mention anything about AmigaOneXE, I said a "GHz class PPC running altivec optimised code".

Quote:
Now you even want me to write your benchmark. You are kidding.


For an alleged hardware engineer, logic doesn't appear to be your strongest point, does it? No, I simply ask you to prove your claim, which is what I have been asking since it came up. I have only clarified, in what I hope are no uncertain terms, the specifics of the test code to make sure it was completely understandable for anyone wanting to have a go.

Quote:
And when I do this then you claim that I cheated and coded the PPC code slow on purpose.
I think I know your style now...


That is very unlikely since I stopped asking you to write a PPC version several posts ago already. Let someone else implement that and you can't be accused of anything. You can both gist your solutions on GitHub and anyone else with appropriate hardware can validate.

I suppose it's also very unlikely because you won't (or perhaps can't since I doubt you are actually Gunnar anyway by this point).

Quote:
Before this lowers down in just a waste of time, how about we make this a little more interesting?

How about we bet some real money on this?
How about we bet $200 or maybe more if you like?
What do you think?


Ooh, a wager! How original. Let me think about it... Erm.... no, No, because I am not an idiot that gambles on something of zero consequence. No, because I am not the one making any claim that I need to put a stake on, rather I am asking you to prove yours.

For the avoidance of all doubt, I have zero investment, financially, emotionally or in any other quantifiable measure, in the outcome whatsoever.

If the 68080 can beat a GHz PowerPC at the proposed vector optimised alpha blending task under ideal conditions for each system, then that's cool AF and kudos. If it can't, well that doesn't bother me either way and I don't think it detracts from the 68080 either.

I was actually interested to see the numbers but as it's clear that it's not going to happen, we may as well move on.

Last edited by Karlos on 05-Oct-2022 at 07:44 PM.

_________________
Doing stupid things for fun...

 Status: Offline
Profile     Report this post  
Gunnar 
Re: Packed Versus Planar: FIGHT
Posted on 5-Oct-2022 20:05:16
#436 ]
Regular Member
Joined: 25-Sep-2022
Posts: 471
From: Unknown

@Karlos
Quote:

I was actually interested to see the numbers


Sorry but what confuses me here is that you do ask for numbers that you already have.
But maybe I was not explaining you this well enough?

What you ask for is nothing else but a memory benchmark.
The AMMX code will runs with memory speed.
And the memory benchmark result you have already!
The number that you ask for is the same number that BUSTEST will give you.
https://aminet.net/package/util/moni/bustest
As you ask for time passes its of course the inverse of the BUSTEST result.

stream, minibench and bustest are three benchmarks that I gave you already which will spit this number.


Honestly beating the AmigaOne XE, is nothing proud off here. This is like kicking a guy a wheelchair.
The Vampire has in all memory benchmarks already beaten the A1XE by a huge margin.
Frankly I find it really puzzling that you even question the result?
That the result is memory bound is crystal clear - therefore the result is known to everyone already.
I think its question of logic to know the result in advance, that you question it does in fact confuse me a little.



Will you come at A37 next week?
Worlds biggest Amiga event?

We can look at the number together there ;)

Last edited by Gunnar on 05-Oct-2022 at 08:17 PM.

 Status: Offline
Profile     Report this post  
cdimauro 
Re: Packed Versus Planar: FIGHT
Posted on 6-Oct-2022 5:06:32
#437 ]
Elite Member
Joined: 29-Oct-2012
Posts: 3621
From: Germany

@Gunnar

Quote:

Gunnar wrote:
Cesare Di Mauro,

Quote:

[quote]
Correct. That's why you can talk about efficiency and then AMMX is clearly more efficient for some operations.

But it's when you talk about performances and saying that AMMX is faster than a PowerPC then you're lying.


Please help us understand what lying means for you.[/quote]
Sure, no problem: see below.
Quote:
Does a CPU with AMMX exist today? Yes!
Can you buy AMMX systems today? Yes!
Do people already have such systems today? Yes! Over 10,000 systems sold
Is AMMX used by several coders? Yes!
Do any programs use AMMX today? Yes!
Does the Operating System using AMMX today? Yes!
Can you play games using AMMX today? Yes!

Those are totally irrelevant: just padding stuff which has no sense regarding my sentence that you've quoted.

Is it lying? Not yet: it's "just" a tentative of derailing the focus of the discussion hiding it on a wall-of-text.
Quote:
Are general graphic routines shorter and simpler to code with AMMX than with ALTIVEC? Yes!
Are game routines shorter and simpler to code with AMMX than with ALTIVEC? Yes!
Do game routines using AMMX need less clock cycles than ALTIVEC? Yes!
Do game routines using AMMX need less memory bandwidth than ALTIVEC? Yes!

Here's the laying part: from my GENERAL statement about AMMX vs PowerPC you start going to specific use cases or details which are totally irrelevant compared to my original statement.

Let me report here again it:

But it's when you talk about performances and saying that AMMX is faster than a PowerPC then you're lying.

As it's possible to understand by the average Joe, I was GENERICALLY talking about PERFORMANCES of AMMX vs PowerPC.

So, how short and simple is some specific routine... does NOT matter.

It does NOT matter it's simpler to code.

It does NOT matter if it needs less clock cycles (it's called "efficiency", and I've ALREADY talked about it before).

And finally it does NOT matter if it needs less bandwidth (again, it's called "efficiency", and I've ALREADY talked about it before).

Everything that you reported it NOT pertinent to my GENERAL statement. But I'll add something more at the bottom to further clarify it.
Quote:
Clearly in all regards AMMX has many advantages!
+ Easier to code
+ Runs faster clock by clock
+ Makes more efficient use of available memory bandwidth

Same as above: it does NOT matter.

This is about EFFICIECY whereas I've clearly talked about PERFORMANCE on the specific statement that you quoted.

Do you understand the difference?

And yes: this is, AGAIN, another LOGICAL FALLACY. Because you're miserably trying to change an argument (PERFORMANCE) with ANOTHER ONE (EFFICIENCY).

But that you're THE King of logical fallacies it's something which is well known...
Quote:
Cesare Di Mauro,
You claimed here several times that you create a much better architecture than Intel.

Does a CPU with your vapor architecture exist today? No!
Can you buy a system your vapor architecture today? No!
Do any programmers code for your vapor architecture today? No!
Does any Operating system support your vapor architecture today? No!
Can you play games on your vapor architecture today? No!
Does a CPU with your vapor architecture reach any clock rate at all today? No!
Can your at least proof that your architecture would even work? No!

AGAIN: another LOGICAL FALLACY.

Specifically, you're trying to DISCREDIT your interlocutor with OTHER things which are absolutely NOT pertinent to the specific discussion.

What's even worse, is that you're inventing complete non-sense as part of your discredit propaganda.

So, TWO logical fallacies... combined together. You really miss nothing!

Anyway, I've already replied here: https://amigaworld.net/modules/newbb/viewtopic.php?mode=viewtopic&topic_id=44169&forum=17&start=0&viewmode=flat&order=0
Quote:
What is your definition of lying?

See above.

And it's enough to see what lying means and your systematic attitude to change the discussion from its original context to try to bring consensus over your, and only your, ideas.

 Status: Offline
Profile     Report this post  
cdimauro 
Re: Packed Versus Planar: FIGHT
Posted on 6-Oct-2022 5:10:29
#438 ]
Elite Member
Joined: 29-Oct-2012
Posts: 3621
From: Germany

@Karlos

Quote:

Karlos wrote:
@Gunnar

Those are bold claims. Your workloads must be very selective. Alpha blending is a good example though. Suppose I have two large pixel, e.g 1080p arrays of ARGB 32-bit pixels and I want to alpha blend buffer B onto buffer A using B's alpha channel.

Are you claiming the 68080, at it's normal clock rate, using AMMX will complete this in less time than a 1GHz PPC using altivec instructions to perform this task?

But those are still synthetic benchmarks.

As I've said before:

It's important to test an entire application / game and not just single routines.

Last edited by cdimauro on 06-Oct-2022 at 05:11 AM.

 Status: Offline
Profile     Report this post  
cdimauro 
Re: Packed Versus Planar: FIGHT
Posted on 6-Oct-2022 5:11:37
#439 ]
Elite Member
Joined: 29-Oct-2012
Posts: 3621
From: Germany

@kolla

Quote:

kolla wrote:
@cdimauro

Quote:

Anyway, the important message was that they were upset with the very bad work done (Again! According to them) by those demo makers.


Hyperbole, jargon and timing suggests tongue-in-cheek humour.

I don't think so, since they had to fix the demo.

That's my impression.

 Status: Offline
Profile     Report this post  
Gunnar 
Re: Packed Versus Planar: FIGHT
Posted on 6-Oct-2022 7:43:50
#440 ]
Regular Member
Joined: 25-Sep-2022
Posts: 471
From: Unknown

@cdimauro

Dear Cesare Di Mauro,

Quote:

Let me report here again it:
But it's when you talk about performances and saying that AMMX is faster than a PowerPC then you're lying.

As it's possible to understand by the average Joe, I was GENERICALLY talking about PERFORMANCES of AMMX vs PowerPC.

So, how short and simple is some specific routine... does NOT matter.

It does NOT matter it's simpler to code.

It does NOT matter if it needs less clock cycles (it's called "efficiency", and I've ALREADY talked about it before).

And finally it does NOT matter if it needs less bandwidth (again, it's called "efficiency", and I've ALREADY talked about it before).

Everything that you reported it NOT pertinent to my GENERAL statement. But I'll add something more at the bottom to further clarify it.
Quote:
Clearly in all regards AMMX has many advantages!
+ Easier to code
+ Runs faster clock by clock
+ Makes more efficient use of available memory bandwidth

Same as above: it does NOT matter.

This is about EFFICIECY whereas I've clearly talked about PERFORMANCE on the specific statement that you quoted.


We can explain this to you.
I will give some real example to help you understand what we talk about here.


Lets have a look at real games and how they are coded.

Lets look the game DIABLO,
Lets look at Command and Conquer,
Lets look at the new RTG version of SONIC the Hedgehog,
Lets look at ROBIN HOOD (MorphOS game)
Lets look at 194x Deluxe
Lets look at Apollo Invader and Apollo-X


All of these games are RTG games for Amiga that came out in last years.
They are real world examples of games that came out Amiga.

Diablo uses 256 color, the other games use 16bit or even truecolor.
Sonic is 320 resolution, the others user 640 or even higher resolution 800/960.

Some of the games are based on SDL ports of PC games.
For example ROBIN HOOD, which I ported to MorphOS is based on SDL.
But also SONIC, DIABLO and COMMAND & CONQUER are based on SDL PC versions.

Some games like SONIC, 194x, Apollo-X use Dual or Multiplayfield effects.
DUAL / Multiplayfield on RTG is done by copying the playfield on top of each other.
This means that the playfields are coded like "huge" Softsprites.

ROBIN HOOD uses many houses and castle or animated elements like Windmills.

Most of the games have hundreds of Sprites on the Screen.
Some of the game like Apollo-Invader, Robin-Hood, Apollo-X, and 194X use alpha blending and light effects.

The internal coding of these games is all similar in many ways.
All these game are coded by using the CPU to create the screen.
This is normal coding style for PC ports, for 2D SDL games and for Amiga RTG Games.

To make these games run you need enough CPU power, and you need to copy a lot of memory.
In other words the performance of all of these games is limited by the memory performance.
This is also the reason why ROBIN HOOD was to a slower on PEGASOS 1 compared to PEGASOS 2 , as the Bus was slower on the Pegasos 1.

You see this also when playing DIABLO on AmigaOne or Pegasos, and compare this with DIABLO on Vampire. The game runs much faster on Vampire.
This is because of 3 reasons
a) The memory is faster on the Vampire
b) the G3 and G4 PowerPC CPU do not support automatic memory prefetching, but the 68080 CPU does
c) You need 3 memory access on the PowerPC - while only 2 memory access on the 68080 with AMMX for doing GFX combines.




Let us explain this with numbers:

The memory on the Vampire give you ballpark 600 MB/sec speed

For the AmigaOne I need someone to run Stream/Minibench again exact number
but from recalling I think we speak about 100-150 MB /sec
I would be nice if someone could run minibench or BUSTEST and please correct me if I recall incorrectly.


For the sake of argument lets say 150 MB
As you can clearly see the Vampire starts with a 4 times memory performance advantage.
600 MB versus 150 MB
This is a huge difference, and the main reason why these type of games run faster on the Vampire.


Now on the POWERPC you need 3 memory operations to process a screen element.
This means to do things like rendering a Sprite, a bullet, and animated background piece, a dual playfield == you always need 3 memory access.


This means you have to divide your memory bandwidth by 3
150 MB / 3 == 50 MB
The AmigaOne has at the end of the day a Sprite/Render/Playfield/Background construct performance of 50 MB/sec

You can do good games with 50MB/sec performance.
But its limiting you a lot in possible resolutions, number of sprites and frame rate.


Lets look at the Vampire.

Because AMMX is more efficient you can do render a Sprite, a bullet, and animated background piece, a dual playfield = with 2 Memory access.

This means you have to divide your memory bandwidth only by 2
600 MB / 2 == 300 MB

The Vampire has at the end of the day a Sprite/Render/Playfield/Background construct performance of 300 MB/sec


As you see this comparison started with a 4:1 memory performance advantage
and with higher efficiency of AMMX over ALTIVEC this ends with a total 6:1 performance advantage


Lets clarify a few things:
a) We all love PowerPC.
I did design parts of some of the best PowerPC cores.
I wrote tons of Altivec performance code for IBM. And I ported and wrote software for MorphOS.

b) The above games are typical examples of RTG games that came to Amiga systems in the last years. There are also other games of course. But I picked the above games because I wrote them or ported them myself or was involved in it, so I know for all of these games perfectly how they are working.




 Status: Offline
Profile     Report this post  
Goto page ( Previous Page 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 | 15 | 16 | 17 | 18 | 19 | 20 | 21 | 22 | 23 | 24 | 25 | 26 | 27 | 28 | 29 | 30 | 31 | 32 | 33 | 34 | 35 | 36 | 37 | 38 | 39 | 40 | 41 | 42 Next Page )

[ home ][ about us ][ privacy ] [ forums ][ classifieds ] [ links ][ news archive ] [ link to us ][ user account ]
Copyright (C) 2000 - 2019 Amigaworld.net.
Amigaworld.net was originally founded by David Doyle