English Amiga Board

English Amiga Board (https://eab.abime.net/index.php)
-   support.WinUAE (https://eab.abime.net/forumdisplay.php?f=5)
-   -   faster winuae ?? (https://eab.abime.net/showthread.php?t=95257)

turrican3 26 November 2018 08:10

faster winuae ??
 
Hi Toni,
do you have some projects to make winuae even faster ???
Is it possible and how ???
What will be the more important feature update in a near future ???
Which feature are you the more proud of ??

DamienD 26 November 2018 11:58

Quote:

Originally Posted by turrican3 (Post 1287270)
Hi Toni,
do you have some projects to make winuae even faster ???
Is it possible and how ???

What do you mean by "make winuae even faster"; in what respect?

meynaf 26 November 2018 12:08

Code optimization maybe ?
Wake up, it's a Windows program, and the natural evolution of such programs is to become slower and heavier as version number rises :D

Toni Wilen 26 November 2018 13:22

How to get it faster:

1) buy a new PC.
2) find someone who wants to rewrite JIT.

I am not that interested in max speed, compatibility is much more important and interesting.

Romanujan 26 November 2018 14:03

Quote:

Originally Posted by Toni Wilen (Post 1287297)
1) buy a new PC.


Did not work for me. Unfortunately, x86 CPU single core performance didn't see any meaningful boost since ages.

But, honestly, why do we need even more speed?

Retro-Nerd 26 November 2018 14:13

Quote:

I am not that interested in max speed, compatibility is much more important and interesting.

Exactly. This should be always the main goal in every emulator or FPGA. :great

Toni Wilen 26 November 2018 14:26

Quote:

Originally Posted by Romanujan (Post 1287301)
Did not work for me. Unfortunately, x86 CPU single core performance didn't see any meaningful boost since ages.

Actually it has because caches have become larger and faster and especially RAM has become faster. This is very important in emulators, there is never enough memory bandwidth.

Normal benchmarks rarely show the difference.

EDIT: btw, don't use sysinfo to measure speed, please :)

EDIT2: JIT most likely generates (at least some) code that today's highly optimized out of order processors can't run optimally.

Romanujan 26 November 2018 14:50

Not using JIT (I'm not happy to pay the price in compatibility for extra speed I don't need). Checked with AIBB (and FS-UAE). Speed about 30% down after moved from Core i7 4790k to Ryzen 2950x... but I don't really care, code compilation just flies now :)

Toni Wilen 26 November 2018 14:56

Quote:

Originally Posted by Romanujan (Post 1287307)
Not using JIT. Checked with AIBB (and FS-UAE). Speed about 30% down after moved from Core i7 4790k to Ryzen 2950x... but I don't really care, code compilation just flies now :)

I read that Haswell added something that helped especially emulators. Ryzen probably isn't as optimal. So in other words, you didn't buy a better PC for emulation :)

Seriously, I upgraded to Ryzen 2700X because I wanted more true cores for compilers and virtual machines and it also lowered WinUAE performance compared to previous 6700K. Not that it can be seen in normal usage.

meynaf 26 November 2018 15:08

IIRC they added something to help indirect calls (aka polymorphism).

Toni Wilen 26 November 2018 16:12

Quote:

Originally Posted by meynaf (Post 1287310)
IIRC they added something to help indirect calls (aka polymorphism).

That would explain the difference. Every CPU memory access requires one indirect call.

meynaf 26 November 2018 17:37

Quote:

Originally Posted by Toni Wilen (Post 1287318)
That would explain the difference. Every CPU memory access requires one indirect call.

Ouch. These are real performance killers on modern cpus, even if the situation is slightly better on some of them...
However, while it's easy to understand why one has to be made to execute an instruction in non-jit mode, wouldn't it be possible to do otherwise for memory accesses ? What is the reason it had to work this way ?

AC/DC HACKER! 26 November 2018 18:51

Quote:

Originally Posted by Toni Wilen (Post 1287305)
EDIT: btw, don't use sysinfo to measure speed, please :)

Awwww, c'mon, it's fun to see that bar waaaaay over the right. Heh, heh! :) :p

Toni Wilen 26 November 2018 20:56

Quote:

Originally Posted by meynaf (Post 1287332)
Ouch. These are real performance killers on modern cpus, even if the situation is slightly better on some of them...
However, while it's easy to understand why one has to be made to execute an instruction in non-jit mode, wouldn't it be possible to do otherwise for memory accesses ? What is the reason it had to work this way ?

I don't think there is no other way than function pointers to support different memory/IO types and sizes dynamically.

turrican3 27 November 2018 09:13

i thought about multi processor... I know that it seems a dead end, but i wanted to know if toni totally abandonned this path and i think it's what he did. But perhaps he keeps an eye on the possibility.
The big difficulty, if you use multi processor, is syncing all amiga chipsets ???
In clear : multi processor will be perhaps a possibility or it will never work enough sufficiencly to use it ???
multi-cpu ,In theory, seems a good way to speed up the emulation but it seems that is bad idea finally... But what could change this ??? More cpu cache, more bandwidth... What could change to make winuae possibly using multi cpu ?
ps: thank you about advice about what we have to look if we change our pc.

Quote:

Originally Posted by meynaf (Post 1287288)
Code optimization maybe ?
Wake up, it's a Windows program, and the natural evolution of such programs is to become slower and heavier as version number rises :D

Meynaf, what is for you the faster winuae version ??

Quote:

Originally Posted by Toni Wilen (Post 1287305)
EDIT: btw, don't use sysinfo to measure speed, please :)

What we should use ???

meynaf 27 November 2018 09:49

Quote:

Originally Posted by turrican3 (Post 1287420)
Meynaf, what is for you the faster winuae version ??

I haven't compared so i can't tell.

AMIGASYSTEM 27 November 2018 10:24

Quote:

Originally Posted by turrican3 (Post 1287420)
Meynaf, what is for you the faster winuae version ??

There are no faster versions, I did a little test calculating the loading speed of my AfA-OS system and all versions 3.x and 4.x have employed the same time, only versions 2.8.1 and 3.0.0 have taken 1 second less.

WinUAE represents the fastest OS3 ever existed and this is one reason why we should not stop and continue to be faster, this in my opinion will stimulate developers to do something new and more modern.

jotd 27 November 2018 10:39

if you're complaining about WinUAE speed because compilations are slow, I suggest that you move to a cross assembler/compiler.

turrican3 27 November 2018 15:49

what is faster : an hdf (pfs) or a simple windows directory ???
To use with dosboxaga ??
edited: Toni, could it be possible to add a setting to make a kind of half speed vsync you know : 50 hz:25fps 60hz:30fps ... It could help to have a not too bad sync for low end pc... Many games have this option now could be interesting to test it.

AMIGASYSTEM 27 November 2018 16:40

From my experience there is not much difference between Directory Filesystem and a HardFile, the difference can be seen instead if you use a real HardDisk; better still on Windows if you use a Hardisk created with the RAM Disk


All times are GMT +2. The time now is 17:04.

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2024, vBulletin Solutions Inc.

Page generated in 0.08241 seconds with 11 queries