sorear changed the topic of #riscv to: RISC-V instruction set architecture | https://riscv.org | Logs: https://libera.irclog.whitequark.org/riscv | Matrix: #riscv:catircservices.org
beber_ has joined #riscv
sajattack has quit [Ping timeout: 268 seconds]
emdevt has joined #riscv
jacklsw has joined #riscv
_whitelogger has joined #riscv
jacklsw has quit [Quit: Back to the real life]
vagrantc has quit [Ping timeout: 276 seconds]
sajattack has joined #riscv
squelch22 has joined #riscv
squelch21 has quit [Ping timeout: 252 seconds]
squelch22 is now known as squelch21
BootLayer has joined #riscv
sajattack_ has joined #riscv
sajattack has quit [Read error: Connection reset by peer]
sajattack_ is now known as sajattack
sajattack has quit [Remote host closed the connection]
squelch22 has joined #riscv
squelch21 has quit [Ping timeout: 276 seconds]
sajattack has joined #riscv
squelch21 has joined #riscv
squelch21 has quit [Client Quit]
squelch22 has quit [Ping timeout: 245 seconds]
_whitelogger has joined #riscv
pjw has quit [Ping timeout: 265 seconds]
pjw has joined #riscv
ray has quit [Quit: ZNC 1.9.0+deb2build3 - https://znc.in]
ray has joined #riscv
mithro has quit [Ping timeout: 272 seconds]
Stat_headcrabbed has joined #riscv
agentcasey_ has joined #riscv
agentcasey has quit [Ping timeout: 252 seconds]
mithro has joined #riscv
haritz has quit [Quit: ZNC 1.8.2+deb3.1+deb12u1 - https://znc.in]
paulbarker has quit [Ping timeout: 245 seconds]
paulbarker has joined #riscv
vagrantc has joined #riscv
vagrantc has quit [Quit: leaving]
jjido has joined #riscv
jacklsw has joined #riscv
jjido has quit [Quit: My laptop has gone to sleep. ZZZzzz…]
jacklsw has quit [Ping timeout: 248 seconds]
emdevt has quit [Remote host closed the connection]
<ganboing> drmpeg: can you describe the performance regression in detail? ref: https://forums.sifive.com/t/low-1-core-stream-bandwidth/7274/13
<drmpeg> It's a real-time application. If a core goes over 100% utilization, then the entire flow loses real-time.
<drmpeg> Let me capture a htop.
<drmpeg> So here it is without the prefetch change. The lpdc_bb7 thread consumed 71.4% for that snapshot. https://www.w6rz.net/atschtop.png
vigneshr has joined #riscv
<drmpeg> With the prefetch change, that thread goes over 100%.
<ganboing> Can I just run it? Does it require any special peripheral?
<drmpeg> Yes, it can run stand-alone.
<drmpeg> I have some older code that I can post. Hang on.
Stat_headcrabbed has quit [Remote host closed the connection]
<drmpeg> This is older code of DVB-T2. You'll have to modify it for ATSC 3.0, which is what I'm running on the P550.
<ganboing> I have a modified openocd that can directly change CSR 7c3 and 7c4, so I can just do a A/B comparison without flashing
<drmpeg> Cool.
<ganboing> What's the fastest way to replicate your results on p550? Build https://github.com/drmpeg/gr-atsc3 and then?
<drmpeg> You can't build gr-atsc3 without GNU Radio.
<drmpeg> It's a big matzo ball.
<drmpeg> If you can wait a day or two, I can make something for you.
<drmpeg> That's tested.
<ganboing> Great, take your time. It's a pretty interesting case. Apparently HW prefetcher settings can significantly impact certain type of workloads
<drmpeg> Yeah, I was surprised.
nexR has joined #riscv
jjido has joined #riscv
kepstin has quit [Remote host closed the connection]
kepstin has joined #riscv
_whitelogger has joined #riscv
Andre_Z has joined #riscv
TMM has quit [Quit: https://quassel-irc.org - Chat comfortably. Anywhere.]
TMM has joined #riscv
coldfeet has joined #riscv
tgamblin has joined #riscv
jjido has quit [Quit: My laptop has gone to sleep. ZZZzzz…]
BootLayer has quit [Quit: Leaving]
jjido has joined #riscv
psydroid2 has joined #riscv
jjido has quit [Quit: My laptop has gone to sleep. ZZZzzz…]
tgamblin has quit [Ping timeout: 240 seconds]
BootLayer has joined #riscv
jedesa has quit [Remote host closed the connection]
jjido has joined #riscv
Armand|5800x has joined #riscv
jacklsw has joined #riscv
jjido has quit [Quit: My laptop has gone to sleep. ZZZzzz…]
jjido has joined #riscv
jacklsw has quit [Quit: Back to the real world]
tgamblin has joined #riscv
Andre_Z has quit [Quit: Leaving.]
TMM has quit [Quit: https://quassel-irc.org - Chat comfortably. Anywhere.]
TMM has joined #riscv
tgamblin has quit [Quit: WeeChat 4.6.3]
coldfeet has quit [Quit: Lost terminal]
jjido has quit [Quit: My laptop has gone to sleep. ZZZzzz…]
Abstract-Wizard has joined #riscv
vagrantc has joined #riscv
coldfeet has joined #riscv
Armand|5800x has quit [Quit: Leaving]
BootLayer has quit [Quit: Leaving]
coldfeet has quit [Remote host closed the connection]
haritz has joined #riscv
haritz has quit [Changing host]
haritz has joined #riscv
psydroid2 has quit [Quit: KVIrc 5.2.6 Quasar http://www.kvirc.net/]
jjido has joined #riscv
ZLima12 has quit [Remote host closed the connection]
ZLima12 has joined #riscv
<drmpeg> ganboing: Here's the test code.
<ganboing> WoW, so fast. Let me try it
<drmpeg> g++ -O2 ldpcatsc3.cc -o ldpcatsc3
<drmpeg> It's basically a giant XOR machine.
<drmpeg> But I think it's the interleavers after the XOR that are slow.
<drmpeg> If you change line 144 to C6_15, it will run the other algorithm that doesn't have interleavers. It's about 3X faster.
<ganboing> So if I see the thread's using 100% CPU, it's busted, right?
<drmpeg> No, this is non real-time.
<drmpeg> You should just see it run slower with the new prefetch values.
jjido has quit [Quit: My laptop has gone to sleep. ZZZzzz…]
<ganboing> I see huge difference. I use openocd/jtag to set CPU3 prefetcher with old values from last release, and CPU2 has the new values
<ganboing> Here's the results:
<ganboing> taskset -c 2 ./ldpcatsc3
<ganboing> average ns = 3989
<ganboing> taskset -c 3 ./ldpcatsc3
<ganboing> average ns = 1933
<ganboing> If I'm interpreting it correctly, performance is halved with the new release
<drmpeg> Yes.