|
The_Franz posted:ARMv8 and i think POWER9 are designed around strong ordering too. herb sutter's atomic weapons talk is pretty interesting since it covers the history of the C++/C11 atomics and gives lots of examples as to why you shouldn't use the relaxed orderings (because even the people who designed this stuff got it wrong most of the time). According to the ARMv8-A guide: "The ARMv8 architecture employs a weakly-ordered model of memory." and "Reads and writes to Normal memory can be re-ordered by hardware, being subject only to data dependencies and explicit memory barrier instructions." There are support for load-acquire and store-release instructions, but that doesn't mean that the ISA itself guarantees strong ordering for regular memory operations. ldra and strl make it easier to write atomic primitives (because you don't need to add extra fences in; the instructions themselves have fence semantics), but the ISA itself still only guarantees weak ordering. I haven't found anything in IBM's documentation that implies that their memory ordering models are any different than the garbage they've used in the past. Of course, IBM"s manuals are so bad that the first author on that paper basically got his Ph.D. just for trying to formally define the POWER memory model. edit: fuckin god dammit
|
# ? Dec 11, 2016 00:14 |
|
|
# ? Jun 12, 2024 20:46 |
|
The Management posted:nope. ARMv8 added ldrl/stra instructions which are sequentially consistent i thought i read something about POWER9 adding sequentially consistent loads and stores without needing heavyweight syncs, but i'm not certain and didn't look too hard into it since i have to reason to touch it. The_Franz fucked around with this message at 00:46 on Dec 11, 2016 |
# ? Dec 11, 2016 00:42 |
|
Menacer posted:edit: fuckin god dammit good hustle
|
# ? Dec 11, 2016 00:44 |
|
The_Franz posted:ARMv8 added ldrl/stra instructions which are sequentially consistent you mean ldra/strl. they are sequentially consistent because they contain half barriers. they are extremely slow compared to the non-barrier versions. on most arm microarchitectures ldra would kill speculation and strl would force waiting on every store getting to the poc.
|
# ? Dec 11, 2016 00:52 |
|
The Management posted:you mean ldra/strl. they are sequentially consistent because they contain half barriers. they are extremely slow compared to the non-barrier versions. on most arm microarchitectures ldra would kill speculation and strl would force waiting on every store getting to the poc. they are still better and cheaper than using a heavyweight barrier around loads and stores when you want sequentially consistent atomic operations like the old arm chips needed. they're not meant to replace regular loads and stores, but they give another compelling reason for not messing with bug-prone relaxed atomic memory ordering anymore.
|
# ? Dec 11, 2016 01:26 |
|
The_Franz posted:they are still better and cheaper than using a heavyweight barrier around loads and stores when you want sequentially consistent atomic operations like the old arm chips needed. they're not meant to replace regular loads and stores, but they give another compelling reason for not messing with bug-prone relaxed atomic memory ordering anymore. they have the same cost as doing a dmb ld or st, respectively. they are also not atomic, you would need to use the exclusive versions (ldrex, strex) if you want to simulate atomicity.
|
# ? Dec 11, 2016 01:41 |
|
The Management posted:they have the same cost as doing a dmb ld or st, respectively. they are also not atomic, you would need to use the exclusive versions (ldrex, strex) if you want to simulate atomicity. they're still potentially more efficient, especially in an out-of-order architecture, since they use acquire-release semantics vs a full barrier which further restricts instruction reordering edit: sequentially consistent stores on arm v7 incur the wrath of two full barriers since the compiler emits dmb st dmb The_Franz fucked around with this message at 04:35 on Dec 11, 2016 |
# ? Dec 11, 2016 02:21 |
|
The_Franz posted:they're still potentially more efficient, especially in an out-of-order architecture, since they use acquire-release semantics vs a full barrier which further restricts instruction reordering armv8 has independent half barriers. dmb ld and dmb st.
|
# ? Dec 11, 2016 17:15 |
|
That sounds like some fairly dmb st, op
|
# ? Dec 11, 2016 19:58 |
|
The surface book just ran an ad featuring a literal manchild as the target market lmfao for loving ever
|
# ? Dec 12, 2016 01:46 |
|
itt we make very serious posts while misusing the term "sequentially consistent"
|
# ? Dec 12, 2016 02:27 |
|
Captain Foo posted:The surface book just ran an ad featuring a literal manchild as the target market
|
# ? Dec 12, 2016 02:39 |
|
Captain Foo posted:The surface book just ran an ad featuring a literal manchild as the target market lol the one where theyre drawing a day of the dead character and he opens with "i made a promise never to grow up" and then every time the woman tries to explain any feature he interrupts with exclamations like "YOU CAN DRAW RIGHT ON THE SCREEN!!!!" like a five year old it was on during football yesterday and my girlfriend said "i thought they threw all those surfaces in the trash" referring to belichicks rant about how lovely they were
|
# ? Dec 12, 2016 13:35 |
|
LastInLine posted:lol the one where theyre drawing a day of the dead character and he opens with "i made a promise never to grow up" and then every time the woman tries to explain any feature he interrupts with exclamations like "YOU CAN DRAW RIGHT ON THE SCREEN!!!!" like a five year old yep it was exactly this
|
# ? Dec 12, 2016 14:35 |
|
You guys are going to love what the UPS truck is bringing me today. https://www.youtube.com/watch?v=_QLsr_jR5gs
|
# ? Dec 12, 2016 16:37 |
|
The Management posted:Microsoft intends to fail in the ARM ecosystem again lol the office team strikes again
|
# ? Dec 12, 2016 19:06 |
|
iyg, of course, is super excited so they can have One Device To Rule Them All and refuse to acknowledge that it'll be worse at any given task than even having two devices
|
# ? Dec 12, 2016 19:16 |
|
apparently dhcp was broken in windows 10 for a while because a service was crashing, and microsoft put a "please reboot your computer" banner at the top of their website about it
|
# ? Dec 13, 2016 21:00 |
|
dhcp: beta tested in the future
|
# ? Dec 13, 2016 21:08 |
|
anthonypants posted:apparently dhcp was broken in windows 10 for a while because a service was crashing, and microsoft put a "please reboot your computer" banner at the top of their website about it holy gently caress, so that's what that was. jfc, how do you break dhcp in an update that goes live and cannot be declined or avoided by a significant number of customers? this is getting to win98 levels of bullshit, except you no longer have the option of never updating at home.
|
# ? Dec 13, 2016 23:10 |
|
i like the part where it says (not shut down) because "shut down" in w10 is some stupid hybrid sleep garbage that doesn't actually reset the operating system that really messed me up for a while because my weird olde habit is to always use shut down instead of restart when i'm doing things that require restarts. feels like i need to give the computer a chance to cool down for 30 seconds before booting it back up again yknow. but anyway that kind of "restart" doesn't count as a restart for any purpose that requires a real restart since microsoft started this poo poo Sagebrush fucked around with this message at 23:51 on Dec 13, 2016 |
# ? Dec 13, 2016 23:45 |
|
infernal machines posted:holy gently caress, so that's what that was.
|
# ? Dec 13, 2016 23:47 |
|
Sagebrush posted:i like the part where it says (not shut down) because "shut down" in w10 is some stupid hybrid sleep garbage that doesn't actually reset the operating system
|
# ? Dec 14, 2016 00:20 |
|
infernal machines posted:jfc, how do you break dhcp
|
# ? Dec 14, 2016 00:20 |
|
Sagebrush posted:feels like i need to give the computer a chance to cool down for 30 seconds before booting it back up again yknow. thats the autism
|
# ? Dec 14, 2016 00:21 |
|
PC users have been so ingrained with "SHUTDOWN" for so many years that they are almost impossible to retrain to put the computer in sleep mode which will save them a bunch of time not going through full boot cycles all the time. so MS completely changed "shutdown" to basically mean write the contents of memory of hiberfil.sys but don't bother retaining the memory contents in ram (S4) so when the computer starts up it pulls everything out of hiberfil which gets you to the desktop quicker than a cold boot. it was... not the best idea.
|
# ? Dec 14, 2016 00:22 |
|
the shutdown function on my operating system (apple macos sierra 10.12.1) works as expected
|
# ? Dec 14, 2016 00:25 |
|
id like to report however, despite its perfect functional record, it is rarely used because i experience long and sustained periods of uptime and stability
|
# ? Dec 14, 2016 00:26 |
|
BangersInMyKnickers posted:PC users have been so ingrained with "SHUTDOWN" for so many years that they are almost impossible to retrain to put the computer in sleep mode which will save them a bunch of time not going through full boot cycles all the time. so MS completely changed "shutdown" to basically mean write the contents of memory of hiberfil.sys but don't bother retaining the memory contents in ram (S4) so when the computer starts up it pulls everything out of hiberfil which gets you to the desktop quicker than a cold boot. it was... not the best idea. now it just hibernates the kernel tho
|
# ? Dec 14, 2016 00:36 |
|
ha ha more like microsucks
|
# ? Dec 14, 2016 00:42 |
|
pram posted:thats the autism i think it's a mild obsessive tendency. not bad enough to call OCD or have a negative effect on my life, but it's there. like sometimes i'll bang my foot on the edge of a door and then my body feels annoyingly unbalanced until i bang my other foot on a door as well. also i have to wash my hands after touching a student's computer but that's just because students are somehow all greasy all the time BangersInMyKnickers posted:PC users have been so ingrained with "SHUTDOWN" for so many years that they are almost impossible to retrain to put the computer in sleep mode which will save them a bunch of time not going through full boot cycles all the time. so MS completely changed "shutdown" to basically mean write the contents of memory of hiberfil.sys but don't bother retaining the memory contents in ram (S4) so when the computer starts up it pulls everything out of hiberfil which gets you to the desktop quicker than a cold boot. it was... not the best idea.
|
# ? Dec 14, 2016 00:43 |
|
"the computer just needs a rest" - a definitely not crazy person
|
# ? Dec 14, 2016 00:56 |
|
you think you're so smart but i bet you've never even heard of capacitor holdup or RAM ghosts jeez
|
# ? Dec 14, 2016 01:54 |
|
i was in the habit of shutting down my computer because printer & graphics tablet drivers would randomly cease to work in windows if i didn't, and i'd have to restart anyway took about 2-3 years of using a mac before i broke out of the habit
|
# ? Dec 14, 2016 02:21 |
|
i know there's a rich tradition of anthropomorphizing complex systems, it's still oogida boogida bullshit you nut.
|
# ? Dec 14, 2016 02:22 |
|
Sagebrush posted:i like the part where it says (not shut down) because "shut down" in w10 is some stupid hybrid sleep garbage that doesn't actually reset the operating system that's not a thing. shut down shuts down. you were hitting sleep
|
# ? Dec 14, 2016 02:22 |
|
wait 30 sec to let the caps discharge before you turn it back on oh is it tired?
|
# ? Dec 14, 2016 02:24 |
|
~*hesitates before pressing the softly pulsing power button, lest the machine become grumpy after waking from sleep*~
|
# ? Dec 14, 2016 02:27 |
|
Shaggar posted:that's not a thing. shut down shuts down. you were hitting sleep I thought the same thing but he's right, shutdown doesn't reinitialize the kernel just suspends it to disk. only restart or hard crash gives you a clean boot. https://www.tenforums.com/tutorials/4189-fast-startup-turn-off-windows-10-a.html BangersInMyKnickers fucked around with this message at 02:32 on Dec 14, 2016 |
# ? Dec 14, 2016 02:27 |
|
|
# ? Jun 12, 2024 20:46 |
|
~*presents a cornucopia of fatty meats and nuts on the altar before rousing the pc from hibernation*~
|
# ? Dec 14, 2016 02:28 |