Overview Features Coding ApolloOS Performance Forum Downloads Products Order Contact

Welcome to the Apollo Forum

This forum is for people interested in the APOLLO CPU.
Please read the forum usage manual.
Please visit our Apollo-Discord Server for support.



All TopicsNewsPerformanceGamesDemosApolloVampireAROSWorkbenchATARIReleases
Performance and Benchmark Results!

Apollo Hardfpu Resultspage  1 2 3 4 5 6 7 8 

Peter Heginbotham

Posts 214
11 Oct 2017 10:15


People are going to be running out of superlative's at this rate. :)


Nixus Minimax

Posts 416
11 Oct 2017 10:24


Markus B wrote:

OK, so it means even better results can be expected from the V4 FPU?

Sure. Even when running at the same clock rate, the V4 FPU will have far more FPU instructions in hardware. The V2 FPU we are currently seeing has only the most frequently used FPU instructions in hardware while the rest is still emulated. The V2 FPU is similar to an 040/060 FPU but trimmed down even more (the 882 being the reference) and already with the relevant parts of the 040/060.library included in the core. So depending on the mix of instructions used in a specific program, the V2 FPU will show varying speed while the V4 FPU can blast through all the code at maximum speed.



Samuel Devulder

Posts 248
11 Oct 2017 10:38


(no POVray results yet.)
   
By the way, is it possible to have technical insight (Jari?) on how improvements in speed of the SoftCore is achieved.
   
Is it that FEmu for the 080 now uses more and more hardware primitives (fadd, fmul..) or maybe enven more primitive like barrel-shifters, adders to implement higher-level functions (fsqrt, flogn, fetox, fsin, fsinh, ...) or some other tricks like the ones in fdlibm ( EXTERNAL LINK ) which does everything using integer operations.


Jari Eskelinen

Posts 23
12 Oct 2017 06:44


Samuel Devulder wrote:

  By the way, is it possible to have technical insight (Jari?) on how improvements in speed of the SoftCore is achieved.
     
  Is it that FEmu for the 080 now uses more and more hardware primitives (fadd, fmul..) or maybe enven more primitive like barrel-shifters, adders to implement higher-level functions (fsqrt, flogn, fetox, fsin, fsinh, ...) or some other tricks like the ones in fdlibm ( EXTERNAL LINK ) which does everything using integer operations.
 

 
  Since my name was mentioned... Credit needs to go where credit belongs. Speeds boost of femu after 0.10 is not my doing, it's combined efforts of multiple people from the team.
 
  You see, like I told earlier, femu was and is my first venture to asm coding. So I learned everything from scratch and even though I got femu to run first mandelbrot generator relatively early, code was complete mess and even I knew it. Slowly I learned more bout asm coding and was able to improve performance a lot by e.g. using jump tables instead of hideous if-then-else structures, using bitfield operations instead of multiple bitwise operations etc. All the stuff veteran asm coders know by heart - but I did not.
 
  But this improved performance only so far. Since I had no asm experience, my code was highly sequential. All instructions depending results of previous ones. This hinders performance a lot since features like pipelining and super scalar did not work like they should. It would have taken me years to learn how to code super scalar friendly, but luckily team members put a lot of effort into this and hand optimized some of my code and instantly tripled the performance(!).
 
  But of course this too goes only so far. Thos insane numbers you see were due to new CPU features. The team kindly showed me how to use them, implemented CPU features to femu in addition to me etc. They did put huge effort to help me with femu and they put huge effort to improve femu themselves. Don't like to talk what these features are since it's work in progress, but you can guess a lot.
 
  So like I said, it was collaborative effort of everybody. Even people who did not code, put a lot of effort into testing etc. I thank them all, amazing job, really amazing. Only Amiga makes it possible, right? ;)
 


Samuel Devulder

Posts 248
12 Oct 2017 11:28


Thank you Jari :D
 
So if I understand correctly, there's a lot of integer computation involved indeed, integer ops that Vampire's gurus optimized by hand to benefit from superscalar architecture as much as possible.


Nixus Minimax

Posts 416
12 Oct 2017 12:11


@Jari:

You did what no Amiga developer did in the last 32 years. Even if there was a lot of room for improvement, you did not hide behind your lack of knowledge or experience in asm coding. You did what Commodore should have done when they published the A3000 at the very latest: a software-FPU would have set an end to the quirky math library solution that required a whole library call per FPU-instruction even for Amigas that did have an FPU. Compilers would simply have generated FPU code and programs would not have guru'ed just for a printf() for a fractional number. I have a lot of respect for what you did. At the rate you have proven to conquer the realm of assembly coding, I'm sure that superscalar programming and other optimisation techniques will become second nature for you in no time. There really isn't that much to good assembly coding as the self-proclaimed asm gurus want to make people believe...



Lord Aga
(Apollo Team Member)
Posts 119
12 Oct 2017 18:57


Jari is a true Amiga hero :)

He didn't whine for months about the FPU, he sat down and did it.
And remained modest. Some people could learn a lot from this.


Manuel Jesus

Posts 155
25 Oct 2017 22:28


To see the FPU progress "from scratch" within the last few months has been amazing.... much more progress than a .1 release every 10 years as others do in the Amiga scene. It was lots of fun bringing up testing TV paint, Lightwave, and Quake FP versions. There is a 3d FPU based real time object viewer I want to tinker with this weekend.


D Lasher

Posts 8
27 Oct 2017 06:54


Lord Aga wrote:

Jari is a true Amiga hero :)
 
  He didn't whine for months about the FPU, he sat down and did it.
  And remained modest. Some people could learn a lot from this.

+1, well done.


Thierry Atheist

Posts 644
27 Oct 2017 17:41


Peter Heginbotham wrote:
People are going to be running out of superlative's at this rate. :)

Hi Peter!!!!

YES.... ABSOLUTELY AGREE TO THAT!!!!!!

Jari Eskelinen wrote:
Since my name was mentioned... Credit needs to go where credit belongs. Speeds boost of femu after 0.10 is not my doing, it's combined efforts of multiple people from the team.

Hi Jari,

THANK YOU for you efforts... THANK YOU for everything that you've put yourself through and done for the AMIGA cause!!!!!

Jari Eskelinen wrote:
Don't like to talk what these features are since it's work in progress, but you can guess a lot.

Yes... yes. It SHOULD be kept SECRET!!!!! (NO to OPEN SOURCE!!!)
Jari Eskelinen wrote:
So like I said, it was collaborative effort of everybody. Even people who did not code, put a lot of effort into testing etc. I thank them all, amazing job, really amazing.

  AND BIG THANK YOUS ALL AROUND TO THOSE THAT I DON'T KNOW, TOO!!!!
Jari Eskelinen wrote:
Only Amiga makes it possible, right? ;)

I've been waiting to get THIS AMIGA FOREVER (<--pun intended) :-D :-D :-D

So yeah, TOTALLY TOO TRUE!!!!

Amiga + Apollo (+NatAmi--->) Vampire F.T.W.!!!!!!!!!

(p.s. It'll be interesting to see what new things will come about with a new assembler programmer (Jari) joining into the Amiga movement.... :-) :-) :-))


John William

Posts 563
28 Oct 2017 00:10


-- Modified --

Thierry Atheist wrote:

Peter Heginbotham wrote:
People are going to be running out of superlative's at this rate. :)

  Hi Peter!!!!
 
I agree.
 
 
Jari Eskelinen wrote:
Since my name was mentioned... Credit needs to go where credit belongs. Speeds boost of femu after 0.10 is not my doing, it's combined efforts of multiple people from the team.

  Hi Jari,
 
    Thank you for all your efforts and thank you for what you have brought to the Amiga community. :)

 
Jari Eskelinen wrote:
Don't like to talk what these features are since it's work in progress, but you can guess a lot.

  I agree that it should be a secret and not be open source.

 
Jari Eskelinen wrote:
So like I said, it was collaborative effort of everybody. Even people who did not code, put a lot of effort into testing etc. I thank them all, amazing job, really amazing.

  Big thanks to all the team who have helped made this possible! <- Single ! is more than enough.

 
Jari Eskelinen wrote:
Only Amiga makes it possible, right? ;)

 
I have being waiting for a long time for something like this to happen! Thank you everyone for all your hard work!

After I modified all the !! and shouting and capitalization I was able to actually read what you said and not enter into maniac ADD and skip everything and not be able to read what you said.

The problem I have is when there is so much shouting and splash and noise in the paragraph I cannot read them, I enter into super, super, super ADD and I just skip and skip and skip and end up asking someone what you wrote above.



Thierry Atheist

Posts 644
28 Oct 2017 06:50


Hee, hee, hee....

Hi John William,

google has got it's work cut out for it, updating "translate" to cope with my posts, eh? :-D

Or, maybe a plug-in for IBrowse, AWeb, Voyager, chrome, Opera, Firefox...


John William

Posts 563
28 Oct 2017 19:01


Thierry Atheist wrote:

Hee, hee, hee....
 
  Hi John William,
 
  google has got it's work cut out for it, updating "translate" to cope with my posts, eh? :-D
 
  Or, maybe a plug-in for IBrowse, AWeb, Voyager, chrome, Opera, Firefox...

Not really. I just slammed on the table over 8,000 pages of the most thickest book title Translating Thierry Atheist Into English.

It toke 24 hours straight to translate just that paragraph!


Simo Koivukoski
(Apollo Team Member)
Posts 601
30 Oct 2017 22:25





Thomas Blatt

Posts 200
31 Oct 2017 04:36


@Simo which Vampire is it?


Simo Koivukoski
(Apollo Team Member)
Posts 601
31 Oct 2017 05:14


Thomas Blatt wrote:

@Simo which Vampire is it?

V600V2@92/Gold2.7


Thomas Blatt

Posts 200
31 Oct 2017 05:23


Awesome:)


Michal Warzecha

Posts 209
31 Oct 2017 08:41


It's jump from 5 to almost 80 Mflops in so short time. Great work Jari and team. Awesome :)

posts 158page  1 2 3 4 5 6 7 8