Home freenode/#haskell: Logs Calendar

Logs: freenode/#haskell

←Prev  Next→
Page 1 .. 451 452 453 454 455 456 457 458 459 460 461 .. 5022
— 502,152 events total
2020-10-06 02:18:59 <hololeap> how important is it to know where to add INLINE pragmas these days? will compiling with -O2 on the latest GHC be sufficient without worrying about using INLINE?
2020-10-06 02:19:26 × solonarv quits (~solonarv@astrasbourg-653-1-252-112.w92-161.abo.wanadoo.fr) (Ping timeout: 272 seconds)
2020-10-06 02:19:31 hololeap is actually using GHC 8.8.4, not the latest one
2020-10-06 02:20:22 <dolio> Depends on the program.
2020-10-06 02:20:42 → berberman_ joins (~berberman@2408:8207:2560:ee60::a44)
2020-10-06 02:20:46 → solonarv joins (~solonarv@adijon-655-1-70-207.w90-13.abo.wanadoo.fr)
2020-10-06 02:23:19 <hololeap> can you elaborate, dolio?
2020-10-06 02:23:42 <dolio> Well, if I remove all the inline pragmas from vector-algorithms, it will be a lot slower.
2020-10-06 02:23:57 <dolio> (Unless I compensate in other ways.)
2020-10-06 02:24:07 × berberman quits (~berberman@2408:8207:2560:ee60:584e:a9ff:fe9b:d3fe) (Ping timeout: 244 seconds)
2020-10-06 02:24:28 → cosimone joins (~cosimone@2001:b07:ae5:db26:f68c:50ff:fe0b:b774)
2020-10-06 02:24:34 × cosimone quits (~cosimone@2001:b07:ae5:db26:f68c:50ff:fe0b:b774) (Client Quit)
2020-10-06 02:24:36 <dsal> If you need something inlined, or need it not inlined, you should say so.
2020-10-06 02:25:26 → cosimone joins (~cosimone@2001:b07:ae5:db26:f68c:50ff:fe0b:b774)
2020-10-06 02:25:30 <dolio> That isn't the only significant factor, necessarily. But sometimes it enables important optimizations, and GHC will not automatically inline everything.
2020-10-06 02:25:55 <dolio> Inlining everything isn't even really desirable.
2020-10-06 02:25:58 → jle` joins (~mstksg@cpe-23-240-75-236.socal.res.rr.com)
2020-10-06 02:25:59 × jle` quits (~mstksg@cpe-23-240-75-236.socal.res.rr.com) (Changing host)
2020-10-06 02:25:59 → jle` joins (~mstksg@unaffiliated/mstksg)
2020-10-06 02:26:10 × cosimone quits (~cosimone@2001:b07:ae5:db26:f68c:50ff:fe0b:b774) (Client Quit)
2020-10-06 02:26:32 <hololeap> i don't understand what makes a function needing to be inlined, where GHC wouldn't already do it automatically. is there a good primer on all this?
2020-10-06 02:26:33 <dsal> Does GHC ever automatically inline anything? I've had pretty big optimizations from telling it to inline things I thought would be obvious.
2020-10-06 02:26:56 <dolio> Well, here's an example from vector-algorithms again.
2020-10-06 02:27:11 <dolio> The algorithms need to be inlined so that the code can be specialized to the comparison function.
2020-10-06 02:27:18 <dsal> If you have a tiny function you use in a lot of places, copying the instructions will be faster than making a function call.
2020-10-06 02:27:33 <dolio> Because if that doesn't happen, the overhead of comparing integers will be massive.
2020-10-06 02:28:01 → brandly joins (~brandly@c-73-68-15-46.hsd1.ma.comcast.net)
2020-10-06 02:28:06 <hololeap> dsal: what determines the size of a function?
2020-10-06 02:28:09 <dolio> Like, a couple instructions on unboxed values vs. function calls on boxed values.
2020-10-06 02:29:19 <dsal> hololeap: I mean like, you write the abstraction that makes sense, but if you have something that's called enough that the overhead of calling it makes a difference in your runtime, inlining might help.
2020-10-06 02:29:57 <dolio> The actual size of the expression is the size.
2020-10-06 02:30:05 <dolio> I'm not sure what exactly the threshold is.
2020-10-06 02:30:39 <dsal> If it's a lot of code, and you use it in a lot of places, inlining while make your code a lot bigger and your cache less effective.
2020-10-06 02:31:20 <dolio> I'm pretty sure `f x = x + x` is small enough for GHC to decide to inline it automatically. But if you're in doubt, and it matters, it's better to annotate it.
2020-10-06 02:31:31 × jle` quits (~mstksg@unaffiliated/mstksg) (Ping timeout: 256 seconds)
2020-10-06 02:31:36 → user2 joins (~user@068-190-177-201.res.spectrum.com)
2020-10-06 02:31:49 <hololeap> ok, it's becoming clearer
2020-10-06 02:32:33 × user2 quits (~user@068-190-177-201.res.spectrum.com) (Client Quit)
2020-10-06 02:33:40 <dolio> I've been experimenting with doing vector-algorithms with backpack, and that doesn't require inlining tons of code.
2020-10-06 02:33:52 <hololeap> the problem is, i've been spending all my time learning haskell focusing on semantics and getting stuff to compile. i haven't looked into performance or tweaking GHC at all beyond using -O2 every time
2020-10-06 02:33:55 <dolio> Because you can specialize the implementation to the comparson/types.
2020-10-06 02:34:42 <dolio> Well, it might not matter a whole lot depending on what you're doing.
2020-10-06 02:35:37 <dolio> If you're trying to beat C's qsort, though, it matters.
2020-10-06 02:36:44 <hololeap> i'm using a lot of newtype wrappers around monad transformer stacks; packing and unpacking while using inherited typeclasses. and a lot of production code i've seen uses INLINE a lot for this type of thing. but i wonder how much of it is necessary.
2020-10-06 02:37:18 → zoom55AA joins (~zoom55aa@068-190-177-201.res.spectrum.com)
2020-10-06 02:37:24 <dsal> Profiling is usually a good step before taking action.
2020-10-06 02:37:45 <dsal> benchmarking helps both as well, but benchmarking as an effort can be pretty misleading.
2020-10-06 02:38:27 <hololeap> i don't even understand the difference between profiling and benchmarking :/
2020-10-06 02:39:18 <dsal> benchmarking is writing a tool that gives you a repeatable measurement.
2020-10-06 02:39:25 <dsal> profiling is using a tool that tells you where time is spent.
2020-10-06 02:39:41 × rusua_ quits (uid124537@gateway/web/irccloud.com/x-mpizelreyaaubxup) (Quit: Connection closed for inactivity)
2020-10-06 02:40:03 <dsal> So profiling might tell you that a particular function is where you spend a lot of your time. So you write a benchmark for that function, and then do stuff to the implementation until the benchmark gives you consistently better results.
2020-10-06 02:40:18 <dsal> It's a little tricky in haskell, but the tools are helpful.
2020-10-06 02:40:30 <hololeap> what tools should i be looking into?
2020-10-06 02:40:53 <dsal> criterion (or whatever that lighter version is people use, but I still use criterion, I think)
2020-10-06 02:41:11 <dsal> And like, wherever "how to profile haskell" takes you.
2020-10-06 02:42:02 <hololeap> cool, i will crawl the web for it, but i also appreciate specific advice from people who have done it.
2020-10-06 02:42:24 <dsal> stack test --profile
2020-10-06 02:42:37 <dsal> It's a report you read.
2020-10-06 02:43:21 <dsal> Your test or application or whatever runs and then you look at it, stroke your bear, say "hmm... I see" and then your code is faster.
2020-10-06 02:44:01 <dsal> In any case, you don't try to optimize anything you're not measuring.
2020-10-06 02:44:13 → nbloomf joins (~nbloomf@2600:1700:83e0:1f40:71a4:5e3f:3433:7ae1)
2020-10-06 02:44:19 <dsal> I've seen every optimization you can think of slow code down.
2020-10-06 02:44:41 → abhixec joins (~abhixec@c-67-169-141-95.hsd1.ca.comcast.net)
2020-10-06 02:44:41 <hololeap> is the species of bear a crucial variable in this equation?
2020-10-06 02:45:02 × cr3 quits (~cr3@192-222-143-195.qc.cable.ebox.net) (Quit: leaving)
2020-10-06 02:45:16 <dsal> Maybe. It's best to go in without assumptions.
2020-10-06 02:45:17 ski . o O ( "Let's hug a bear." )
2020-10-06 02:45:57 → falafel joins (~falafel@2605:e000:1527:d491:99fe:5613:f0a7:56f0)
2020-10-06 02:46:17 → Stanley00 joins (~stanley00@unaffiliated/stanley00)
2020-10-06 02:47:55 × ralejs_ quits (~ralejs@2620:10d:c093:400::5:2842) (Read error: Connection reset by peer)
2020-10-06 02:48:23 → ralejs joins (~ralejs@2620:10d:c093:400::5:2842)
2020-10-06 02:49:31 → vicfred joins (~vicfred@unaffiliated/vicfred)
2020-10-06 02:50:51 ← zoom55AA parts (~zoom55aa@068-190-177-201.res.spectrum.com) ("WeeChat 2.9")
2020-10-06 02:51:40 → zoom84 joins (~zoom55aa@068-190-177-201.res.spectrum.com)
2020-10-06 02:51:47 × lagothrix quits (~lagothrix@unaffiliated/lagothrix) (Killed (hitchcock.freenode.net (Nickname regained by services)))
2020-10-06 02:51:55 → lagothrix joins (~lagothrix@unaffiliated/lagothrix)
2020-10-06 02:54:06 × falafel quits (~falafel@2605:e000:1527:d491:99fe:5613:f0a7:56f0) (Remote host closed the connection)
2020-10-06 02:56:58 hackage reanimate-svg 0.12.0.0 - SVG file loader and serializer https://hackage.haskell.org/package/reanimate-svg-0.12.0.0 (DavidHimmelstrup)
2020-10-06 02:58:35 → jle` joins (~mstksg@unaffiliated/mstksg)
2020-10-06 02:58:54 → falafel joins (~falafel@2605:e000:1527:d491:99fe:5613:f0a7:56f0)
2020-10-06 03:00:01 × rob01 quits (~rob0@178.162.204.214) ()
2020-10-06 03:01:34 → drbean joins (~drbean@TC210-63-209-99.static.apol.com.tw)
2020-10-06 03:02:55 × berberman_ quits (~berberman@2408:8207:2560:ee60::a44) (Ping timeout: 240 seconds)
2020-10-06 03:03:40 × jle` quits (~mstksg@unaffiliated/mstksg) (Ping timeout: 256 seconds)
2020-10-06 03:04:09 → berberman joins (~berberman@2408:8207:2563:5ae0:584e:a9ff:fe9b:d3fe)
2020-10-06 03:06:25 × drbean quits (~drbean@TC210-63-209-99.static.apol.com.tw) (Client Quit)
2020-10-06 03:16:08 → notzmv` joins (~user@177.45.26.174)
2020-10-06 03:17:23 × notzmv quits (~user@unaffiliated/zmv) (Ping timeout: 240 seconds)
2020-10-06 03:17:27 → elliott_ joins (~elliott@pool-108-51-141-12.washdc.fios.verizon.net)
2020-10-06 03:17:38 × bloodstalker quits (~bloodstal@46.166.187.178) (Read error: Connection reset by peer)
2020-10-06 03:18:04 <zoom84> when I load a module with a type error in a function, GHCI shows me the expected and actual types, and also shows me all the bindings. is there a way to display the bindings on command for a given function, outside of compile errors
2020-10-06 03:18:51 <dsal> The bindings? Are you debugging?
2020-10-06 03:18:58 <dsal> Or do you just want type annotations?
2020-10-06 03:18:59 <dsal> :T fix
2020-10-06 03:19:00 <dsal> :t fix
2020-10-06 03:19:02 <lambdabot> (a -> a) -> a
2020-10-06 03:19:21 <zoom84> more exploring than debugging
2020-10-06 03:19:37 <zoom84> :t shows me the sig. i'm looking to the the sig and the bindings that resulted from it
2020-10-06 03:19:39 <lambdabot> error:
←Prev  Next→
Page 1 .. 451 452 453 454 455 456 457 458 459 460 461 .. 5022

All times are in UTC.