Re: SMP on Mac99

[Andrew Randrianasulu]

пт, 28 февр. 2025 г., 21:52 Andrew Randrianasulu <randrianasulu@gmail.com>:

пт, 28 февр. 2025 г., 21:36 Jd Lyons <lyons_dj@yahoo.com>:

Encoding some video will be a true test we can run on OS 9, OS X and linux. With single CPU and dual, but I’m pretty confident from the full results of Geekbench we did what we set out to do.

Usually people do few runs on one config, and few runs on another config, so caching (host side) will not interfere that much...

I diffed 2 "binary" (but really xml?) GeekBench files attached to mail to qemu-ppc and they seems to confirm SMP working?

diff -u binaVyi8GEdr_.bin binal8rPzTzhR.bin

[it will be nice if webinterface to mail list set real filenames on attach download, but for small number of files we can manage rename manually ]

=====

-<result threads="1" simd="0" result="43232449.2" comment="41.2 MB/sec" score="938" percent="9" />

+<result threads="1" simd="0" result="52996014.6" comment="50.5 MB/sec" score="1150" percent="11" />

+<result threads="2" simd="0" result="94817606.1" comment="90.4 MB/sec" score="2206" percent="22" />

 </results>

 </benchmark>

 <benchmark name="Text Compress" id="102" units="2">

 <results>

-<result threads="1" simd="0" result="3424126.0" comment="3.27 MB/sec" score="1021" percent="10" />

+<result threads="1" simd="0" result="4260770.5" comment="4.06 MB/sec" score="1270" percent="12" />

+<result threads="2" simd="0" result="7950853.9" comment="7.58 MB/sec" score="2311" percent="23" /> </results>

 </benchmark>

 <benchmark name="Text Decompress" id="103" units="2">

 <results>

-<result threads="1" simd="0" result="4841381.2" comment="4.62 MB/sec" score="1123" percent="11" />

+<result threads="1" simd="0" result="5961534.4" comment="5.69 MB/sec" score="1383" percent="13" />

+<result threads="2" simd="0" result="11076468.4" comment="10.6 MB/sec" score="2651" percent="26" />

 </results>

 </benchmark>

 <benchmark name="Image Compress" id="104" units="9">

 <results>

-<result threads="1" simd="0" result="9040584.8" comment="9.04 Mpixels/sec" score="1094" percent="10" />

+<result threads="1" simd="0" result="11286957.6" comment="11.3 Mpixels/sec" score="1366" percent="13" />

+<result threads="2" simd="0" result="20489692.1" comment="20.5 Mpixels/sec" score="2435" percent="24" />

 </results>

 </benchmark>

 <benchmark name="Image Decompress" id="105" units="9">

 <results>

-<result threads="1" simd="0" result="10287879.1" comment="10.3 Mpixels/sec" score="612" percent="6" />

+<result threads="1" simd="0" result="11888546.6" comment="11.9 Mpixels/sec" score="708" percent="7" />

+<result threads="2" simd="0" result="19285634.1" comment="19.3 Mpixels/sec" score="1182" percent="11" />

 </results>

 </benchmark>

 <benchmark name="Lua" id="107" units="10">

 <results>

-<result threads="1" simd="0" result="207310.8" comment="207.3 Knodes/sec" score="538" percent="5" />

+<result threads="1" simd="0" result="265836.8" comment="265.8 Knodes/sec" score="690" percent="6" />

+<result threads="2" simd="0" result="482048.0" comment="482.0 Knodes/sec" score="1253" percent="12

=====

Even single thread seems to be a bit faster now?

But I am confused about Debian 12 netinstall iso

one at

https://cdimage.debian.org/cdimage/ports/12.0/powerpc/

360 mb from 2023

I booted it on qemu with Jd's patches, but installer itself uses non-smp kernel and grub install fail.

another

https://cdimage.debian.org/cdimage/ports/current/

450mb, 25 feb 2025.

not tested this one because I pucked one below ....

https://cdimage.debian.org/cdimage/ports/snapshots/2024-12-24/

2024-12-25 08:57

491M

I tried to boot this one and it hanged earlier for me.

What exact url you used for debian-12 iso netinstall test?

Now we just need to shoot for 4 CPUs;-)

On Feb 28, 2025, at 1:20 PM, Howard Spoelstra <hsp.cat7@gmail.com> wrote:

On Fri, Feb 28, 2025 at 7:14 PM Jd Lyons <lyons_dj@yahoo.com> wrote:

The quick and dirty is Geekbench score went from 743 to 1033 Mac OS X Tiger 10.4.11.

So it looks like this actually works and speeds thing up!

I reverted all my changes and just used Balaton’s patches, three for Qemu and one for OpenBois.

Now we just need to patch Openbios to give additional cpus the state of stopped.

The true test for that cpu is to run:

ioreg -n PowerPC,G4@1 

in a terminal and look at its state.

Best,

Howard