NeXT Computers Forum Index NeXT Computers
www.NeXTComputers.org
 
Log in to check your private messagesLog in to check your private messages

Log inLog in  RegisterRegister


Profile  Search  Memberlist  FAQ  Usergroups
What Needs to be done for a NeXT Emulator
Goto page Previous  1, 2, 3 ... 74, 75, 76
 
Post new topic   Reply to topic    NeXT Computers Forum Index -> Emulation / Virtualization
View previous topic :: View next topic  
Author Message
ardi



Joined: 11 Jun 2014
Posts: 84

PostPosted: Sun Jul 16, 2017 8:35 am    Post subject: Reply with quote

Thanks a lot!!! All these changes are great improvements indeed!!

BTW, am I right supposing that the source code is the r833 version from the softfloat branch at https://sourceforge.net/p/previous/code/HEAD/tree/branches/branch_softfloat/ ?

Thanks a lot... maybe after the FP changes, Previous might run on PowerPC Macs now... Laughing
Back to top
View user's profile Send private message
andreas_g



Joined: 30 Jan 2009
Posts: 412
Location: Austria

PostPosted: Sun Jul 16, 2017 9:50 am    Post subject: Reply with quote

ardi wrote:
BTW, am I right supposing that the source code is the r833 version from the softfloat branch at https://sourceforge.net/p/previous/code/HEAD/tree/branches/branch_softfloat/ ?

Yes, that is correct.

ardi wrote:
Thanks a lot... maybe after the FP changes, Previous might run on PowerPC Macs now... Laughing

As long as you don't use the NeXTdimension, yes (but very slow)! Unfortunately the i860 emulation core is not endianness safe. It only works on little endian systems and therefore the NeXTdimension emulation won't run on PowerPC Macs or other big endian systems.
Back to top
View user's profile Send private message Send e-mail
ardi



Joined: 11 Jun 2014
Posts: 84

PostPosted: Sun Jul 16, 2017 1:04 pm    Post subject: Reply with quote

andreas_g wrote:
ardi wrote:
BTW, am I right supposing that the source code is the r833 version from the softfloat branch at https://sourceforge.net/p/previous/code/HEAD/tree/branches/branch_softfloat/ ?

Yes, that is correct.

ardi wrote:
Thanks a lot... maybe after the FP changes, Previous might run on PowerPC Macs now... Laughing

As long as you don't use the NeXTdimension, yes (but very slow)! Unfortunately the i860 emulation core is not endianness safe. It only works on little endian systems and therefore the NeXTdimension emulation won't run on PowerPC Macs or other big endian systems.


Great!! I'm not particularly interested in running it in PowerPC regularly, but it's certainly good to know that I might run it there in case of emergency (although of course I assume the build won't be straightforward, as SDL2 is in the game... but I think that it should build and run with little or moderate effort).
Back to top
View user's profile Send private message
t-rexky



Joined: 09 Jan 2011
Posts: 270
Location: Snowy Canada

PostPosted: Sun Jul 16, 2017 2:42 pm    Post subject: Reply with quote

andreas_g wrote:
it took me longer than expected but finally I was able to finalize Previous v1.7!

This release concentrates mainly on improving accuracy and reliability.


Andreas, this is another SUPERB release of Previous! Thank you so very much for all your efforts!

Previous has never been more stable. I noticed a huge improvement in networking reliability. I also exercised the FPU emulation with my very old strip analysis based propeller geometry optimization code and I can happily report that it works extremely well. While the overall results are not identical to the real world hardware, they are extremely close! The code is iterative and it heavily uses the following functions:

Code:
_acos
_atan
_cos
_pow
_sin
_tan


Awesome work, truly awesome! Thank you!
Back to top
View user's profile Send private message
andreas_g



Joined: 30 Jan 2009
Posts: 412
Location: Austria

PostPosted: Sun Jul 16, 2017 10:53 pm    Post subject: Reply with quote

t-rexky, we did lots of tests on the new FPU code, but I am delighted to hear it works for real world applications!

What real hardware did you run your program for comparison? Was it 68030/68882 or 68040 or something different? NeXTstep 2.0 or any later version?

Normally the results should be a perfect match on 68040, because in fact it does not execute the transcendental functions you mentioned in your post. It does emulate them in software using only simple functions, like FADD, FMUL, etc. Then the difference is probably caused by not 100% identical FSAVE stack frames. Those are a little bit complicated Wink

If you have run your test on 68882/68030 then the difference is easy to explain. Previous' 68882 emulation gives results identical to FPSP (library for 68040 that emulates transcendental functions in software) which does not perfectly match 68882. Emulating 68882 bit perfect would involve reversing its internal CORDIC algorithm. Maybe some fun project for a mathematician, but definitely not trivial.
Back to top
View user's profile Send private message Send e-mail
t-rexky



Joined: 09 Jan 2011
Posts: 270
Location: Snowy Canada

PostPosted: Mon Jul 17, 2017 7:21 am    Post subject: Reply with quote

Hi Andreas, I did the A/B comparison runs on my real TurboColor and on a Previous emulated TurboColor. Both running NS3.3. When I have a moment I will look to see if I can trace the source of the differences...
Back to top
View user's profile Send private message
t-rexky



Joined: 09 Jan 2011
Posts: 270
Location: Snowy Canada

PostPosted: Mon Jul 17, 2017 7:33 pm    Post subject: Reply with quote

@Andreas: root cause identified! It was me. I re-run the optimization code on my TurboColor and on Previous emulated TurboColor and the results are 100% identical, with the exception of the analysis time. Excellent!
Back to top
View user's profile Send private message
andreas_g



Joined: 30 Jan 2009
Posts: 412
Location: Austria

PostPosted: Tue Jul 18, 2017 2:43 am    Post subject: Reply with quote

That is great news! This confirms that our test results have been valid and the FPU emulation is indeed bit-perfect for non-transcendental operations.

btw. even though 68882 transcendental function results are not bit perfect, they are unlikely to cause problems, because they are in fact more precise than the results of a real 68882.
It is documented in the Motorola data sheets that FPSP has same or excess precision compared to 68882 and our emulation exactly matches FPSP.
Back to top
View user's profile Send private message Send e-mail
ardi



Joined: 11 Jun 2014
Posts: 84

PostPosted: Tue Jul 18, 2017 5:41 am    Post subject: Reply with quote

t-rexky wrote:
@Andreas: root cause identified! It was me. I re-run the optimization code on my TurboColor and on Previous emulated TurboColor and the results are 100% identical, with the exception of the analysis time. Excellent!

Just curious, how much different was the analysis time? Do you get the impression that Previous is always faster or always slower than the real hardware, or maybe that the average is well tuned and it's just a matter of an application using more or less instructions that are "overaccelerated"/"overslowed" ? (I don't have real hardware to test, that's the reason why I ask).
Back to top
View user's profile Send private message
eagle



Joined: 29 Mar 2006
Posts: 388

PostPosted: Tue Jul 18, 2017 5:47 am    Post subject: Reply with quote

ardi wrote:
Just curious, how much different was the analysis time? Do you get the impression that Previous is always faster or always slower than the real hardware, or maybe that the average is well tuned and it's just a matter of an application using more or less instructions that are "overaccelerated"/"overslowed" ? (I don't have real hardware to test, that's the reason why I ask).


My gut tells me that it's faster than my 68030. Maybe faster than my 68040-25 too, but then I always emulate a turbo color station and all I have is a non-turbo mono station.
_________________
My NeXTs:
NeXT Computer prototype (68030-25 x2, 68040-25)
Two NeXTstations (68040-25)
All mono
Back to top
View user's profile Send private message
t-rexky



Joined: 09 Jan 2011
Posts: 270
Location: Snowy Canada

PostPosted: Tue Jul 18, 2017 5:50 am    Post subject: Reply with quote

ardi wrote:
Just curious, how much different was the analysis time? Do you get the impression that Previous is always faster or always slower than the real hardware, or maybe that the average is well tuned and it's just a matter of an application using more or less instructions that are "overaccelerated"/"overslowed" ? (I don't have real hardware to test, that's the reason why I ask).


Subjectively with NEXTSTEP 3.3 it feels approximately the same while running on my 21" 2011 iMac (my "kitchen" machine). The optimization time was very close, as in ~10 minutes on real hardware vs. ~9.5 minutes on Previous. This was with Previous emulating 33 MHz 68040. When I set Previous speed to the last option (I think it's titled "variable"), for some reason it appears to get a bit slower, as in >11 minutes...

As a side note, my 2009 MacBook Pro completes the same solution in approximately 1 second. Most of the time is spent in I/O while it's dumping intermediate optimization data to files and status to the console Smile
Back to top
View user's profile Send private message
Nightengale



Joined: 04 Jan 2006
Posts: 148
Location: SD CA

PostPosted: Sat Jul 29, 2017 10:12 am    Post subject: help Reply with quote

Hi community. Can some kind soul please point me to a step by step for dummies like myself to get previous 1.7 running on my macbook? It has os x 10.12.1. So far I have downloaded Previous 1.7, downloaded two roms Rev_3.3_v74.bin and Rev_2.5_v66.bin and the Previous manual. step by step please from doubleclicking the Previous icon, sorry I'm lost with the emulation virtualization scene. I hope I'm not too lazy Very Happy
_________________
NeXT Turbo Color 4.2, HP Pavillion w/4.2,MacBook Pro, PPC G5 dual 2.7 10.5.8.
Back to top
View user's profile Send private message
Kitchen2010



Joined: 26 Feb 2014
Posts: 22

PostPosted: Mon Jul 31, 2017 12:54 pm    Post subject: Reply with quote

I have updated the Wiki entry @"http://com-emu.pixub.com/doku.php/emulator/previous" for the Previous emulator regarding the fixes and changes for the v1.7!

Keep up the great work on this emulator! Razz
Back to top
View user's profile Send private message
itomato



Joined: 27 Dec 2005
Posts: 301
Location: San Francisco, CA

PostPosted: Thu Sep 21, 2017 12:07 am    Post subject: Reply with quote

I built and packaged (vanilla) Previous 1.7 r841 for OS X.

Available here: https://github.com/itomato/Previous-DMG/releases/tag/1.7
_________________
-itomato
Back to top
View user's profile Send private message Visit poster's website
Display posts from previous:   
Post new topic   Reply to topic    NeXT Computers Forum Index -> Emulation / Virtualization All times are GMT - 7 Hours
Goto page Previous  1, 2, 3 ... 74, 75, 76
Page 76 of 76

 
Jump to:  
You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot vote in polls in this forum



Powered by phpBB © 2017 phpBB Group