| + B[] (~Thunderbi@219-89-228-78-adsl.sparkbb.co.nz) | 00:33 | |
| - erlehmann (QUIT: Remote host closed the connection) (~erlehmann@dynamic-046-114-038-246.46.114.pool.telefonica.de) | 01:22 | |
| + erlehmann (~erlehmann@dynamic-046-114-038-246.46.114.pool.telefonica.de) | 01:22 | |
| + jfred_ (quassel@2600:3c03::f03c:92ff:fe35:a0e9) | 03:07 | |
| - devyl (QUIT: *.net *.split) (devylmatri@gateway/shell/matrix.org/x-xlqhwcrmmsuecnrb) | 03:13 | |
| - jfred[m]1 (QUIT: *.net *.split) (jonterracr@gateway/shell/matrix.org/x-oqyrfqdcebokvuum) | 03:13 | |
| - MilkManzJourDadd (QUIT: *.net *.split) (milkmanz1@gateway/shell/matrix.org/x-raqcqalycsmjjsxw) | 03:13 | |
| - thebigfrog (QUIT: *.net *.split) (~thebigfro@51.154.21.134) | 03:13 | |
| - jfred (QUIT: *.net *.split) (quassel@fsf/member/terracrypt) | 03:13 | |
| - kremlin (QUIT: Ping timeout: 246 seconds) (~kremlin@uglyman.kremlin.cc) | 03:17 | |
| + kremlin (~kremlin@uglyman.kremlin.cc) | 03:19 | |
| - chartreuse (QUIT: Quit: Leaving) (~chartreus@S0106f0f249dfd9c3.cg.shawcable.net) | 05:12 | |
| + oomono (uid328183@gateway/web/irccloud.com/x-tghvuujxdofesdyn) | 10:04 | |
| + chomwitt (~chomwitt@2a02:587:dc4f:fd00:c843:19c:a87e:7114) | 10:30 | |
| + jfred[m]1 (jonterracr@gateway/shell/matrix.org/x-oqyrfqdcebokvuum) | 10:35 | |
| + MilkManzJourDadd (milkmanz1@gateway/shell/matrix.org/x-raqcqalycsmjjsxw) | 10:35 | |
| + devyl (devylmatri@gateway/shell/matrix.org/x-xlqhwcrmmsuecnrb) | 10:35 | |
| + thebigfrog (~thebigfro@51.154.21.134) | 10:35 | |
| - B[] (QUIT: Ping timeout: 260 seconds) (~Thunderbi@219-89-228-78-adsl.sparkbb.co.nz) | 11:01 | |
| - ezequielg (QUIT: Read error: Connection reset by peer) (sid363064@gateway/web/irccloud.com/x-ippqujrqjstrmnos) | 12:14 | |
| + ezequielg (sid363064@gateway/web/irccloud.com/x-zexopgqtexublgva) | 12:15 | |
| - kylophone (QUIT: Read error: Connection reset by peer) (sid403812@gateway/web/irccloud.com/x-hmzoydqhkkzucjyv) | 12:15 | |
| - paulk-leonov (QUIT: Ping timeout: 260 seconds) (~paulk-leo@vpn-0-22.aquilenet.fr) | 12:15 | |
| + kylophone (sid403812@gateway/web/irccloud.com/x-myuvabhcjvmykiin) | 12:16 | |
| + paulk-leonov (~paulk-leo@vpn-0-22.aquilenet.fr) | 12:25 | |
| - ehmry (QUIT: Quit: https://quassel-irc.org - Chat comfortably. Anywhere.) (~quassel@2a03:3b40:fe:ab::1) | 12:50 | |
| + ehmry (~quassel@2a03:3b40:fe:ab::1) | 12:51 | |
| - oomono (QUIT: Quit: Connection closed for inactivity) (uid328183@gateway/web/irccloud.com/x-tghvuujxdofesdyn) | 13:43 | |
| - buckket (QUIT: Remote host closed the connection) (~buckket@pdp8.buckket.org) | 15:30 | |
| + buckket (~buckket@pdp8.buckket.org) | 15:35 | |
| - chomwitt (QUIT: Ping timeout: 260 seconds) (~chomwitt@2a02:587:dc4f:fd00:c843:19c:a87e:7114) | 15:55 | |
| FUZxxl | mntmn: didn't you say something about the LS1046A being a possible alternative SoC for the Reform? | 16:39 |
|---|---|---|
| mntmn | FUZxxl: rbz+us are developing a LS1028A SOM | 16:42 |
| FUZxxl | mntmn: Ah, I see. Seems like I have confused the product numbers. | 16:42 |
| FUZxxl | (LS1046A support has landed in FreeBSD HEAD today) | 16:42 |
| mntmn | cool | 16:45 |
| mntmn | the 1028A has the same vivante GC7000L gpu as imx8m | 16:45 |
| mntmn | but it can address more memory and has bigger pcie address space | 16:45 |
| mntmn | it is dual cortex-a72 vs quad cortex-a53 of imx8mq. | 16:45 |
| FUZxxl | that sounds cool | 16:46 |
| FUZxxl | Cortex A72---that's the same core used in the RPi 4B | 16:46 |
| FUZxxl | I've written some SIMD code for that chip in the last weeks | 16:47 |
| FUZxxl | the performance is okay, but not stellar | 16:47 |
| FUZxxl | might also be because the SIMD units have at least a 3 cycle latency on all operations | 16:47 |
| FUZxxl | SIMD performance is about 1/4 the speed of my Haswell-based laptop (using 128 bit vectors on both; if I use AVX on the Haswell box it gets a lot worse) | 16:49 |
| + oomono (uid328183@gateway/web/irccloud.com/x-vwfaejnbcbysaptk) | 17:03 | |
| mntmn | FUZxxl: interesting | 17:28 |
| mntmn | FUZxxl: anything public? | 17:29 |
| FUZxxl | mntmn: https://github.com/clausecker/pospop | 17:30 |
| FUZxxl | you need to build a Go toolchain from master as the code requires some recently added extensions to the assembler to work. | 17:31 |
| FUZxxl | If you build with an older toolchain, the SIMD version won't be built | 17:31 |
| mntmn | FUZxxl: interesting! is the perf tied to caches as well? | 17:32 |
| FUZxxl | mntmn: the ARM64 code doesn't have prefetching added yet | 17:33 |
| FUZxxl | on x86, cache size makes a huge difference | 17:34 |
| FUZxxl | it eventually regresses to being main memory bound | 17:34 |
| mntmn | ok | 17:35 |
| FUZxxl | though that's only after the L3 cache is exhausted | 17:35 |
| FUZxxl | on the Cortex-A72, there doesn't really seem to be an effect | 17:35 |
| FUZxxl | I'm getting the same performance for each buffer size. Perhaps the code is too slow for memory bandwidth to matter? | 17:35 |
| FUZxxl | It only reaches around 3.6 GB/s | 17:36 |
| - bachler (QUIT: Ping timeout: 258 seconds) (~bachler@ip200.dorrhaken1.se) | 17:37 | |
| mntmn | FUZxxl: really interesting. i should run it on the LS1028A and the LX2160A when i have the time | 18:18 |
| FUZxxl | mntmn: cool! You can run the tests with ' go get github.com/clausecker/pospop; go test -bench=. github.com/clausecker/pospop' | 18:21 |
| + B[] (~Thunderbi@219-89-228-78-adsl.sparkbb.co.nz) | 18:27 | |
| - oomono (QUIT: Quit: Connection closed for inactivity) (uid328183@gateway/web/irccloud.com/x-vwfaejnbcbysaptk) | 19:13 | |
| + oomono (uid328183@gateway/web/irccloud.com/x-aiwkihvrtoizwnlx) | 20:55 | |
| - erlehmann (QUIT: Ping timeout: 240 seconds) (~erlehmann@dynamic-046-114-038-246.46.114.pool.telefonica.de) | 21:52 | |
| + erlehmann (~erlehmann@dynamic-046-114-036-100.46.114.pool.telefonica.de) | 22:06 | |
| + chomwitt (~chomwitt@2a02:587:dc4f:fd00:c843:19c:a87e:7114) | 22:41 | |
| - chomwitt (QUIT: Ping timeout: 240 seconds) (~chomwitt@2a02:587:dc4f:fd00:c843:19c:a87e:7114) | 23:29 | |
Generated by irclog2html.py 2.17.3 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!