**** BEGIN LOGGING AT Mon Oct 07 02:59:57 2019 Oct 07 06:28:49 Morning! Oct 07 07:33:44 Morning Oct 07 09:34:30 the new testing build seems to be going on well Oct 07 09:34:53 thought it's nearly 4h per build Oct 07 09:42:03 Yeah, it's slow ;) Oct 07 12:40:28 heh, raspberrypi3-64 and raspberrypi4-64 aren't using the same DEFAULTTUNE after all https://github.com/webOS-ports/meta-webos-ports/commit/677324c83b8dd5d2ed67d071ff0d3b5936a70dca what do you prefer? disable one of them or change them both to use the same DEFAULTTUNE? Oct 07 12:42:57 JaMa: I think performance hit shouldn't be that noticable probably when we use the same? Oct 07 12:43:41 If we can support both at nearly no cost, I'm for it Oct 07 12:44:54 Tofe: What CPU is on the Pine devices? Oct 07 12:45:03 Rockchip RK3399 ? Oct 07 12:46:28 Ah it's Rockchip RK3328 it seems Oct 07 12:46:32 The PineBook uses the 3399. Oct 07 12:46:46 3399 is CortexA72 as well it seems: https://en.wikipedia.org/wiki/ARM_Cortex-A72 Oct 07 12:46:55 RK3328 = CortexA53 Oct 07 12:51:11 "pipeline length or depth: A72 has 15 stages pipeline whereas A53 implements 8 stages pipeline. So, A72 will have, virtually, 15 instruction at the same time in different stages all the time, whereas, A53 will be having only 8 instruction. Accordingly, A72 will be the first to finish executing instructions." Oct 07 12:51:17 That seems quite significant Oct 07 12:51:33 Then again for our current use that extra performance is not really needed Oct 07 12:51:36 Herrie: but we cannot share it with Pine, because of e.g. different mesa, that's why we're using aarch64-cortexa53 only for rpi3-64 now Oct 07 12:52:43 the difference in cores is bigger, but that doesn't mean that the code generated with different -mcpu will actually be different Oct 07 12:53:10 I mean we're not loosing A72 internal improvements when running more generic binaries Oct 07 13:02:05 Yes the execution pipeline should still be more efficient on A72 Oct 07 13:02:21 even if the compiler doesn't optimize the code for that pipeline Oct 07 13:05:02 JaMa: FYI, on pinephone I'm now mostly content with the Mesa SRCREV I'm using. It's still only in master branch, but for next version we should be able to converge Oct 07 13:34:08 the most interesting part is crypto armv8 support enabled in a72, but disabled in a53 DEFAULTTUNE we're currently using => armv8a-crc-crypto would be possible best denominator, but e.g. last comment here https://www.raspberrypi.org/forums/viewtopic.php?t=207888 shows that it might not work correctly on rpi3-64 Oct 07 15:09:43 lets see how this works https://github.com/webOS-ports/meta-webos-ports/commit/058397edecb0a6a2fec991cef6c76d3753a63a9e Oct 07 19:34:04 JaMa: I just realized, but... llvm is useless for pinephone and hammerhead, isn't it ? as I don't intend to have software rasterizer, I use the dedicated gallium driver Oct 07 23:14:50 Tofe: IIRC at least in some mesa versions sw rasterizer was always enabled when gallium was (and build was failing without) so it was mandatory dependency before, I think it was fixed in very recent versions, but don't remember the details, if I don't forget I'll check tomorrow, now almost off to bed **** ENDING LOGGING AT Tue Oct 08 02:59:57 2019