egg|nomz|egg changed the topic of #kspacademia to: https://gist.github.com/pdn4kd/164b9b85435d87afbec0c3a7e69d3e6d | Dogs are cats. Spiders are cat interferometers. | Космизм сегодня! | Document well, for tomorrow you may get mauled by a ネコバス. | <UmbralRaptor> egg|nomz|egg: generally if your eyes are dewing over, that's not the weather. | <ferram4> I shall beat my problems to death with an engineer.
<bofh>
egg|zzz|egg: I'd view it as more of an agreement for review :P
<bofh>
with a positive comment being the analog of a "yes, publish" review
<egg|zzz|egg>
bofh: feature request: DOIs for tweets
<bofh>
I mean, that could legit be useful on nontrivially many occasions tbh
<UmbralRaptop>
bofh: The example would be fine if he got a Taylor series, but whatever that is…
<bofh>
UmbralRaptop: exactly, I'm not even sure what the hell that *is*
<kmath>
<chc40> @sigfig now you can call people dumb for not getting monads while you call people dumb for not getting relativity.… https://t.co/peO2Atu2Up
<kmath>
<kourge> @ursatong @spookperson @FakeUnicode i bring you... the most meta annotation https://t.co/trDJ3UF7WT
<egg|zzz|egg>
bofh: I wonder whether my method is actually correctly rounded; that seems unlikely, but I'm not sure how to check
<egg|zzz|egg>
(I guess since it's probably not correctly rounded, throw millions of inputs at it to get stats? bounding the maximum error seems hard tho)
<bofh>
I *think* you need to do formal analysis if you want to be certain over all inputs, but that's genuinely above my floating-point skill level & you'd have to poke Atlas. Most things just randomize a few million inputs and check what the max error is like, then rerun a few times (w/a different starting seed for your RNG each time).
<egg|zzz|egg>
!wpn UmbralRaptop
* Qboid
gives UmbralRaptop a hypergolic ☣
<bofh>
!wpn egg|zzz|egg
* Qboid
gives egg|zzz|egg a chain clutter
<egg|zzz|egg>
!wpn bofh
* Qboid
gives bofh a hacked bandersnatch
* egg|zzz|egg
should zzz
<egg|zzz|egg>
!wpn whitequark and котя
* Qboid
gives whitequark and котя an avalanche deuterium preta
<egg|zzz|egg>
!wpn the котяchrome kitten
* Qboid
gives the котяchrome kitten a projected induction
<egg|zzz|egg>
!wpn the grey kitten
* Qboid
gives the grey kitten a Shannon constructor
<egg|zzz|egg>
котяchrome seems like a good name :-p
<kmath>
<lukasneville> @hardsci academia: why are we producing so many p-hacked pieces of shit ⏎ academia: why is everything broken into m… https://t.co/PUEnw9ddGc
<UmbralRaptop>
(Also, apparently typical autorejection for many NSF graduate fellowships: GPA < 3.8, don't have a publication)
<UmbralRaptop>
either/or
<UmbralRaptop>
Whee, power went out for several seconds.
<kmath>
<✔bfod> we’re on a weird trajectory. in the end the most popular meme format will be a 12000-word peer-reviewed paper with… https://t.co/0vKfyEALp5
tawny has quit [Ping timeout: 198 seconds]
APlayer has joined #kspacademia
egg|cell|egg has joined #kspacademia
egg|phone|egg has quit [Ping timeout: 198 seconds]
egg|phone|egg has joined #kspacademia
egg|cell|egg has quit [Read error: -0x1: UNKNOWN ERROR CODE (0001)]
egg|cell|egg has joined #kspacademia
egg|phone|egg has quit [Read error: Connection reset by peer]
egg|phone|egg has joined #kspacademia
egg|cell|egg has quit [Read error: Connection reset by peer]
<egg|work|egg>
!wpn whitequark
* Qboid
gives whitequark a plasma gravitational histomorphism
<kmath>
<eggleroy> @stephentyrone Erratum: that was [1, 8[, not [0, 8[.
<whitequark>
!wpn kmath
* Qboid
gives kmath a hypergolic cutting lightning with a cone attachment
<whitequark>
!wpn Qboid
* Qboid
gives Qboid a radio hug
<whitequark>
nice.
<egg|work|egg>
!wpn atlas
* Qboid
gives atlas a vortex
<whitequark>
!wpn GOES-13
* Qboid
gives GOES-13 a mountain
<bofh>
egg|work|egg: whoopsie, lol
<egg|work|egg>
bofh: I wonder whether I can optimize something by moving the coefficients around to require fewer multiplications
<egg|work|egg>
bofh: also there's a multiplication by 64, I should probably do that one on the ALU since I don't care about the behaviour on over/underflow because rescaling
<egg|work|egg>
(not sure whether any of this is useful, unless it's in the critical path it's a waste of time to look at it :-p)
<bofh>
also fp multiplies are cheap
<egg|work|egg>
bofh: yeah it's more if there's somewhere where I'm using up all the throughput I have on FP multiplies
<egg|work|egg>
which might be a possibility on those big polynomials
<egg|work|egg>
that would turn one add into a multiply (the 2y) but it might be better in the end
<bofh>
also keep in mind that if the divisions cause some coefficients to become inexact that might be an overall loss (also a net gain if the opposite happens, tho).
<bofh>
but I extremely doubt that you're wasting throughput on that, by which I mean plz2profiling data :P
<egg|work|egg>
bofh: yeah, and iacatime etc.
<egg|work|egg>
!wpn bofh
* Qboid
gives bofh an unamused catbus
* egg|work|egg
pets the catbus
* bofh
also pets the catbus
* bofh
double-checks for any undocumented code they've written lately, just-in-case
<kmath>
<UnofficialNAM> Question: "what will you do if JWST explodes on launch", speaker manages to calmly say "that would be a bad day in… https://t.co/o0ELabwAgU
<kmath>
<cosmos4u> "Laser-pushed spaceflight concepts, such as 'Breakthrough Starshot', would produce brighter and tighter beams than… https://t.co/PzOUebjpgs
<kmath>
<✔bfod> we’re on a weird trajectory. in the end the most popular meme format will be a 12000-word peer-reviewed paper with… https://t.co/0vKfyEALp5
<egg|zzz|egg>
bofh: um, perhaps I should do the integer stuff with SSE2 intrinsics instead of having the compiler move things back to the r*x registers,
<bofh>
egg|zzz|egg: that won't be faster for scalar stuff, you'd need to actually vectorize it for it to help.
<bofh>
old Athlons may benefit, but that's it. mov xmm, gpr & vice-versa is pretty fast, check Agner Fog's tables :p
<egg|zzz|egg>
for some reason it decides to do it through qword ptr [rsp+0x30]
<egg|zzz|egg>
so that's not great
<egg|zzz|egg>
also it screws with IACA's dependency analysis
<bofh>
ffs, MSVC. it still isn't too bad, but is there an MSVC switch to say "I want this compiled optimized for Sandy Bridge or later?"
<bofh>
like the equivalent of -march=core2 for gcc
<bofh>
also like, does IACA not have a "run this code and tell me what CPU perf counters report" option? static perf analysis is kinda meh tbh.
* Qboid
gives котяchrome kitten a transuranic license
<UmbralRaptop>
Nuclear kitten?
<UmbralRaptop>
"If you know DOS, you know everything"
<UmbralRaptop>
(Density Of States)
<bofh>
I mean, yes.
<egg|zzz|egg>
bofh: yeah compiling the cat's code would involve translating this weird C-processed thing into masm and that sounds unfun
<egg|zzz|egg>
(it involves getting rid of all the macros and comments as well as AT&T -> intel)
<bofh>
egg|zzz|egg: I could do it tonight in the very late evening by hand, that isn't *too* bad (also fuck AT&T syntax so goddamn much).
<egg|zzz|egg>
bofh: I could do it with godbolt maybe >_>
<egg|zzz|egg>
I can do the conversion with godbolt, I'll see if I can do the preprocessing
<egg|zzz|egg>
(is C this preprocessor -> assembler pattern a common thing?)
<bofh>
just do the preprocessing by cpp -E and feed the result into godbolt
<bofh>
the C preprocessor -> assembler pattern is a semi-common thing b/c gas doesn't really have anything approaching a usable assembler macro system, nobody uses it with competent assemblers like [mny]asm, they just use... the assembler's own macro engine.
<egg|zzz|egg>
bofh: I did cpp -E with godbolt :-p
<egg|zzz|egg>
bofh: but doing the gas to masm conversion with godbolt seems less pleasant, the disassembler seems to try to interpret the perfect cube table with entertaining results
<bofh>
just take out the table temporarily. or if you're lazy enclose the whole code in a __asm__ __volatile(" <blarg> "); block
<SnoopJeDi>
Rocky Kolb's talk was pretty good (and very amusing), even for a pleb like me who cannae into cosmological QFT