@Spectre660

Quote:

How well does QEMU emulate an e500v2 SPE ?

Never tried such setup. I used only preconfigured Debian distribution for PowerPC750 (G3 class) CPUs. My skills are too weak to build Linux enviroment, I fear.

As of QEMU emulation, I expect it is the same - softfloats. But this is only my quess.

@pavlor

what do you think?

http://qemu.weilnetz.de/qemu-tech.html

QEMU generic features:

User space only or full system emulation.
Using dynamic translation to native code for reasonable speed.
Working on x86, x86_64 and PowerPC32/64 hosts. Being tested on ARM, HPPA, Sparc32 and Sparc64. Previous versions had some support for Alpha and S390 hosts, but TCG (see below) doesn’t support those yet.
Self-modifying code support.
Precise exceptions support.
Floating point library supporting both full software emulation and native host FPU instructions.

@OlafS25

Not all emulated CPU architectures support all features, slooooow FPU in PowerPC emulation is one of these examples.

@pavlor

I feared so... I did not find specific informations for PPC shortly

@Spectre660

Quote:

Spectre660 wrote: @OlafS25 Powerpc floating point emulation under Winuae is currently very slow too.

Because there's no FPU JIT for it.
Quote:

Winuae 0.3642x with AKsaw.wav Sam440ep is 3.6575x (1:15s) Tabor SPE is 4.0484x (1:09s) (Linux) Sam460ex is 5.8033x (0.47s)

Can you try with the normal PowerPC (no SPE binary)? Also for mplayer. To see the difference between SPE and emulated FPU.

It'll be more interesting to see some Blender result (again, compiled with and without SPE support), because it stresses more the FPU (mplayer is more dependent from the memory subsystem, AFAIK).

@cdimauro

As Spectre660 already wrote, MPlayer doesn´t work in his non-SPE setup.

@cdimauro

Cant currently Install Blender on Debian 8 SPE due to dependencies issues.
Also Mplayer does not work with the video files for the non SPE Linuxes.

Blender Tabor powerpc32
blender -b Documents/blender_projects/bricks_and_water_plugin.blend -o ram:pool -F JPEG -f 0
Saved: ram:pool0000.jpg Time: 10:49.35 (Saving: 00:04.99)

Blender Sam460ex Debian 8
Saved: ram:pool0000.jpg Time: 00:18.01 (Saving: 00:00.18)


Blender results for other machines .
http://amigaworld.net/modules/newbb/viewtopic.php?mode=viewtopic&topic_id=35671&forum=34&start=160&viewmode=flat&order=0#665666


Lame Ubuntu 12.04 powerpc32
Encoding /home/amigaone/Documents/AKsack.wav
to /home/amigaone/Documents/AKsack.mp3
Encoding as 44.1 kHz j-stereo MPEG-1 Layer III (11x) 128 kbps qval=3
Frame | CPU time/estim | REAL time/estim | play/CPU | ETA
10529/10529 (100%)| 1:44:06/ 1:44:06| 1:45:13/ 1:45:13| 0.0440x| 0:00
-------------------------------------------------------------------------------

Quote:

cdimauro wrote: @Spectre660 Quote: Spectre660 wrote: @OlafS25 Powerpc floating point emulation under Winuae is currently very slow too. Because there's no FPU JIT for it. Quote: Winuae 0.3642x with AKsaw.wav Sam440ep is 3.6575x (1:15s) Tabor SPE is 4.0484x (1:09s) (Linux) Sam460ex is 5.8033x (0.47s) Can you try with the normal PowerPC (no SPE binary)? Also for mplayer. To see the difference between SPE and emulated FPU. It'll be more interesting to see some Blender result (again, compiled with and without SPE support), because it stresses more the FPU (mplayer is more dependent from the memory subsystem, AFAIK).

Last edited by Spectre660 on 22-Oct-2015 at 06:52 PM.
Last edited by Spectre660 on 22-Oct-2015 at 05:56 PM.

_________________
Sam460ex : Radeon Rx550 Single slot Video Card : SIL3112 SATA card

@Spectre660

Thanks for your patience, non-SPE lame was painfully slow.

@pavlor

Under Linux you can execute the faster PowerpcSPE programs from a chroot environment at the same times as the regular powerpc32 ones. So you could be running a regular powerpc Qupzilla browser as well as an PowerpcSPE Midori browser at the same time.

Quote:

pavlor wrote: @Spectre660 Thanks for your patience, non-SPE lame was painfully slow.

Last edited by Spectre660 on 22-Oct-2015 at 07:19 PM.

_________________
Sam460ex : Radeon Rx550 Single slot Video Card : SIL3112 SATA card

@Spectre660: thanks for the new tests, which are really interesting.

The LAME test, running on a regular (non SPE: just PowerPC) Ubuntu is 93 times slower than the SPE version, but that's because it doesn't use the SPE unit for the FPU emulation. So, all integer math (only normal PowerPC instructions) is used, which is clearly very slow.

Looking at Blend, the situation is different. Here there isn't an SPE version of the binary, and there is only the result of the SAM460ex. We can safely assume that Tabor/SPE has a similar performance of the 460 (looking at the other results), and we see that it's 36 times slower than the 460ex. So, I guess that here the Debian PowerPCSPE FPU emulations makes use of the SPE unit, internally, getting a 3x gain over the FPU emulation which makes use of the only PowerPC instructions.

So, it's confirmed that the FPU emulation is deadly slow (contrary to the Hyperion statement), but using the SPE version there's a good gain, albeit the timings aren't good anyway.

The only way to make use of Tabor is through the use of ad-hoc compiled binaries for every applications and part of the o.s. which makes "not simple" use of floating point computations.

@cdimauro

Quote:

So, I guess that here the Debian PowerPCSPE FPU emulations makes use of the SPE unit, internally, getting a 3x gain over the FPU emulation which makes use of the only PowerPC instructions.

I fear you assumptions go too far. I doubt non-SPE version of Linux uses SPE unit at all.

As future OS4 "classic" FPU emulation performance is not known yet, we are free to speculate.

@pavlor: please, re-read. I stated that Debian/PowerPCSPE uses the SPE unit, so it has a 3x better experience over Ubuntu/PowerPC, which only uses the regular PowerPC instructions (no SPE used).

@pavlor


It is very much used .

http://amigaworld.net/modules/newbb/viewtopic.php?mode=viewtopic&topic_id=40622&forum=2&start=660&viewmode=flat&order=0#771545


Quote:

pavlor wrote: @cdimauro Quote: So, I guess that here the Debian PowerPCSPE FPU emulations makes use of the SPE unit, internally, getting a 3x gain over the FPU emulation which makes use of the only PowerPC instructions. I fear you assumptions go too far. I doubt non-SPE version of Linux uses SPE unit at all. As future OS4 "classic" FPU emulation performance is not known yet, we are free to speculate.

_________________
Sam460ex : Radeon Rx550 Single slot Video Card : SIL3112 SATA card

@cdimauro

I still don´t get it. WHERE you have Blender SPE result for comparison?

@Spectre660

I was refering to powerpc32 version.

@pavlor

Quote:

pavlor wrote: @cdimauro I still don´t get it. WHERE you have Blender SPE result for comparison?

I've assumed that Blender/SPE = Blender/460ex, since in the other tests Tabor showed similar results. It's just a guess, so no exact results, but just a trend.

@pavlor

The powerpc32 uses traps.
Interestingly I am able to play an MP3 file in the powerpc32 when using Radeon hdmi audio out via a Radeon Hd7750 video card. if usinga Radeon HD 6670 the sound is garbled or plays and stops. No problem with playback with either under powepcspe .


Quote:

pavlor wrote: @Spectre660 I was refering to powerpc32 version.

_________________
Sam460ex : Radeon Rx550 Single slot Video Card : SIL3112 SATA card

@cdimauro

Quote:

I've assumed that Blender/SPE = Blender/460ex, since in the other tests Tabor showed similar results. It's just a guess, so no exact results, but just a trend.

That is what I wrote, we are free to speculate...

I repeat, I doub´t SPE unit is used in powerpc32 Linux for anything. This is my speculation.

@pavlor: but I never stated this. Please, pavlor, re-read my original comment: it's everything clearly explained.

@cdimauro

You clearly defined performance of Hyperion´s FPU emulation (we know nothing about...). Your crystal ball is better than mine.

What we know now about e500v2:
Integer performance
Native FPU performance (SPE)
"Performance" of classic FPU emulated in software

What we don´t know:
Performance of classic FPU emulated on SPE (if possible...)