01:07 imirkin: karolherbst: seems off
01:07 imirkin: 38: export b128 # o[0x0] %r186 %r189 %r192 %r195 (0)
01:07 imirkin: i don't think that's properly restored
01:07 karolherbst: ohhh
01:08 karolherbst: yeah, seems wrong
01:10 karolherbst: mhh weird, the spill/unspill pairs seem to be very wrong
01:11 karolherbst: also the last loads get wrong register assigned
01:14 karolherbst: ahh
01:15 karolherbst: "merge b128 %r265q %r186 %r189 %r192 %r195 (0)" -> merge b128 %r265q
01:15 karolherbst: yeah okay, I am kind of aware of this issue
01:15 karolherbst: ohhh
01:16 karolherbst: okay, it makes sense now
01:16 karolherbst: %186 was spilled, but no unspill was inserted
01:16 karolherbst: and the merges not fixed up
01:19 karolherbst: imirkin: better? https://gist.github.com/karolherbst/460c012ae492db73df88d51859c03f42
01:26 imirkin: cant check now
01:26 karolherbst: okay
01:26 karolherbst: at least those silly "mov u32 $r3 $r1" showed up again, so it can't be that wrong
01:49 mangix: so uhhh
01:49 mangix: what does the inline keyword for a function do exactly?
01:49 mangix: is the point of it to reduce binary size?
01:50 karolherbst: imirkin: okay, found other issues
01:50 karolherbst: mangix: more like the contrary
01:51 mangix: the reason i ask is i've added the keyword to ~10 functions. each reduced the size of the resulting .ko file
01:51 mangix: usualy by 48 bytes
01:53 karolherbst: weird
01:54 karolherbst: usually gcc is smart enough to inline properly
01:54 karolherbst: except you compile with O0 or something
01:54 karolherbst: or don't allow gcc to inline stuff
01:55 mangix: no idea
01:55 mangix: everything is stock on my end
01:56 mangix: adding inline to gk104_gpio_new saved a whopping 112 bytes
01:58 mangix: hrm gcc 7.2.1
01:58 mangix: thought it was supposed to be more aggressive
02:19 karolherbst: there is a kernel config to disable automatic inlining
02:39 mangix: hrm yeah it has to be -O0
03:18 imirkin: mangix: -Os optimizes for size
03:18 imirkin: -O2 optimizes for perf
03:18 imirkin: which one is used depends on some kernel build settings
09:06 mlankhorst: how do I make the tegra power on after power loss?
13:01 mlankhorst: oh looks like I just need to wait a minute, or an hour with official kernel :)
20:37 karolherbst: imirkin: how much do you know about the actual register allocation part of RA?
21:41 imirkin: karolherbst: quite little.