Skip to content
PodSubs
ShowsPeopleGuestsJobsNewsEvents
More
BlogPodCred
Menu
ShowsPeopleGuestsJobsNewsEventsBlogPodCredNetworksSign in
Join nowSign in
  • Shows
  • People
  • Jobs
  • News
  • More
    • Guests
    • Events
    • Blog
    • PodCred
    • Networks
    Join nowSign in

ADSP: Algorithms + Data Structures = Programs

ADSP: Algorithms + Data Structures = Programs artwork

Episode 78: C++ Algorithms & Profiling with Ben Deane (Part 4)

May 20, 202233 minAudio

In this episode, Bryce and Conor finish their conversation with Ben Deane about C++ Algorithms! Twitter ADSP: The Podcast Conor Hoekstra Bryce Adelstein Lelbach About the Guest: For Ben Deane, C++ wasn’t even among the first 10 languages that he learned on his programming journey, but it’s been the one that has paid the bills for the last 20-odd years. He spent most of that time in the games industry; many of the games he worked on used to be fondly remembered but now he’s accepted that they are probably mostly forgotten. These days he works in the finance industry writing high-frequency trading platforms in the most modern C++ that compilers can support. In his spare time he watches a lot of YouTube’s educational sector, practices the Japanese art of tsundoku, reads about the history of programming, avoids doing DIY, and surprises his wife by waking in the middle of the night yelling, “of course, it’s a monad!” before going back to sleep and dreaming of algorithms. Show Notes Date Recorded: 2022-04-19 Date Released: 2022-05-20 ADSP Episode 72: C++ Algorithm Family Feud! ADSP Episode 75: C++ Algorithms with Ben Deane (Part 1) ADSP Episode 76: C++ Algorithms with Ben Deane (Part 2) ADSP Episode 77: C++ Algorithms with Ben Deane (Part 3) quick-bench.com Tyler Weaver Tweet Wordle Guass’s Calendrical Algorithm C++ std::sort C++ std::nth_element C++ std::max_element C++ std::reduce C++ std::transform reduce C++ std::accumulate Google Benchmark Thrust Parallel Algorithms Beautiful Folds (A tutorial on the universality and expressiveness of fold) McNugget Numbers Intro Song Info Miss You by Sarah Jansen https://soundcloud.com/sarahjansenmusic Creative Commons - Attribution 3.0 Unported - CC BY 3.0 Free Download / Stream: http://bit.ly/l-miss-you Music promoted by Audio Library https://youtu.be/iYYxnasvfx8

Audio and video play straight from the show's own hosting. PodSubs stores and serves no media.

Transcript

From the show. Tap a time to play from there.

Ben DeaneThis is reminding me this is Bryce's restaurant algorithm. This is sort of reminiscent of um is it the chicken nuggets algorithm?

ConorWhat come to ADSP the podcast episode 78, recorded on April 19th, 2022. My name is Connor, and today with my co-host Bright, we finish up part four of our four-part conversation with Ben Dean of C algorithms. Don't don't don't worry, uh, we will have you back, Ben, because clearly we bring Ben we bring Ben on and we actually end up talking about what the name of our podcast is, algorithms.

ConorLike we we talked about data structures and algorithms for all of four podcasts, and ones where this is like reduced the podcast, so I don't know what you're talking about.

BryceThat's true. That's true.

ConorI mean, let's let's actually we'll take 30 more seconds as a divergent. Go go to the website, and we have a very nice tagging system where you can go and look at all the different topics and algorithms is one of them.

Ben DeaneThe one that jumped out of me is the life advice topic.

ConorSo we have zero, one, episode 0, 1, 23, 25, 36, 37. Is that the largest by category? 51, 54, 55, 56, 72. Definitely not. C I think is the largest. So that's about 10, 10, or 15 of our 75. So that's not that bad. That's about 20%. The largest has got to be C.

Ben DeaneAnd how many are in life advice? Life advice.

ConorJust one. Episode 44 should be a good one.

BryceCome on, we're giving them a lot more life advice than that.

ConorI mean, I'll be honest, I create these tags a lot of the times while I'm editing the individual podcast and then completely forget about strategic decision making. I think that was part three of the Chandler Patricia podcast.

BryceYes, yes.

ConorWhat is K?

BryceIt's a language.

ConorUh K was the very derivative language. Lowercase. It's the deriv derivative language of K. Yeah.

BryceIf there's anything in the tags that you don't know, there's three episodes.

Ben DeaneIt being lowercase is what threw me.

BryceWe are gonna have to talk more about Wordle because there's an article in the ACCU magazine about Wordle that I haven't gotten to yet. But anyways.

ConorI almost lost last yet. I mean, today's today's Wordle.

BryceI I actually got today's easily.

ConorI got down to my sixth guess, and I could I knew I could think of two words and I knew there was more. And uh I ended up guessing correctly, and I went and did my wordle hindsight. There was four different words I could have guessed. I got I had a 25% chance, so I almost lost my streak of I haven't lost yet. And I I was I was gonna retire today. I thought for sure I was gonna miss it. You were gonna say something about Wordle.

BryceWait, wait, Connor. So once you miss, once you break your streak, you're gonna retire, you're gonna rage quit.

ConorI'm exhausted because I play four different wordles. I play Wordle, Quartle, another Quartle, which is Quantum Wortle.

BryceYou actually play Quantum Wordle?

ConorDo you play Wordle? I've played Wordle once.

BryceYou want to join my quantum compiler team? Because if you can play, if if you can play quantum wordle, then you can join, then you're qualified to be in my quantum compiler team.

ConorI mean quantum mortal is upsetting because it uses a different word list.

BryceAnd so a lot of times That is not the only reason that quantum Wordle is upsetting.

Ben DeaneI'm I'm not that big on Wordle, to be fair. Um, mostly because when it came out, I I immediately thought about it and I thought, well, this is just solvable by computer. Um, so it's not that interesting. Um however, um I've recently been made aware of a you know my world, my wordle strategy now is the ultimate sort of play it safe. Like I don't want to get it into, I just like I just type in uh thanks to my son, he told me this strategy where you go flame, brick, shunt, podgy.

Ben DeaneThey are four five-letter words that have no letters in common, and they cover like once you've typed in those four, you've covered most letters, and then you should be able to get it in two guys.

BryceWhat was what was hang on? Because I like to write down different people's strategies. So tell me, tell me what's the first one.

Ben DeaneWell, this is this is my son Henry's strategy. Flame, brick, shunt, and podgy.

BryceShunt and podgy.

Ben DeaneYeah.

BryceThat's pretty good, that's pretty good coverage. All right, all right. Let's let's Connor. I I want to see the Alright.

ConorWe've we've held this off. Potentially, it's been a week for the listener.

BryceIt feels like it's been a week for me.

ConorZooming out too much. We've got the custom code written by yours truly.

BryceYou're we're gonna have to make a few modifications to it because I hard-coded it in there. But I guess actually it's really just you're gonna have to um uh rename, yeah, change that to V and then change 10,000 to N. And boom, you're done. That's so easy. Oh, you're gonna have to add uh random as an incredible.

Ben DeaneYou want to be um parens there, not braces, right? Because you're making a vector of one element which is n, I think.

BryceOh god! Burns me again! Burns me again.

ConorInitializer list, everyone's favorite thing about C.

BryceI hate it so much. I hate it so much. Random. That's the name of the header, not yeah, random. Yeah.

ConorAlright. Um, we are running the benchmark. Will it compile first? We'll find out in about five seconds how good is Bryce's C.

BryceOh, my C is amazing. Um, yeah, but it's because you don't have C20 enabled.

ConorNo, we we don't we can't use C tad um if we don't have Oh no, we do actually.

BryceWe just we are what's actually what what's the error here?

ConorOh, because we didn't but it doesn't it doesn't know how to CTAD because we don't we didn't give it an element Right right we we switched from initializer list to a constructor specifying the number of elements so there's no actual data to oh C that's why we get paid job security through code obscurity it's my favorite quote from the Scott Myers books is the uh at one point he says job security through code obscurity.

ConorUm do you well this is probably gonna take another 30 seconds. Do you want to just mention that or you want to go? I was gonna mention some more algorithms.

BryceAre you talking at C next?

Ben DeaneNot this year, I decided to take a break and not submit. Um I was gonna mention more favorite algorithms for for future episodes, if you like. Um I am pretty interested in calendrical algorithms, like I was saying to you the other the algorithm for f for finding the date of Easter um is is an algorithm that drove much mathematical progress in the Western world some hundreds of years ago.

ConorIt's interesting the things that cause um technology or mathematics, like you know, the fact that um that gaming is like driven a lot of technological advances. Yeah um and that you yeah, you mentioned that me that to me the other day, Ben, about the Easter, and I had no idea.

Ben DeaneIt's a difficult thing, it turns out, and it took up mathematicians' time in whenever the 16th century of 17th century.

ConorInteresting. Is there a specific name of an algorithm that is used for that, or is it just a general category?

Ben DeaneThe usually like the most understandable algorithm is attributed to Gauss.

ConorInteresting. And the results are back. Bryce completely vindicated about the data mattering because it has completely changed the order. And now what we've said in the past is actually false. Uh the two max element algorithms basically they're not identical because in the lower numbers, um, max element two definitely uh is slower, but then once you get to basically about a thousand elements, they're the same.

BryceThis all of this demonstrates is that we still have insufficient data. This is just from one sequence. The the the way that to actually the way to actually benchmark an like an order-dependent problem like this, you need to run an ensemble where you vary the seed of the input. Um, so like for me to you know find this data believable, one, I'd want to do multiple trials of you know each you know test, but also I would want to do like a hundred trials, each with a different input sequence.

BryceUm uh uh by different input sequence, I mean changing the seed to the random number generator. Um, and I would also want to test a few special sequences. Um, some of the special sequences um would be like the sequence of the input is all the same element. They're all zero. Um and one would be like, you know, reverse sorted or something. Yeah, yeah. Reverse sorted. Um, like, you know, the first two elements happen to be the, you know, the highest, the last two elements happen to be the highest.

BryceUm yeah, so you'd want to try some battery of that because I I bet you if we changed the seed around a few more times, we would see um more variance. Now, the even just this input sequence that we're using is far better than the previous input sequence because the previous input sequence was regular. Um, it would be regular because of this modulo pattern. So the the highest numbers would always be the you know the 50th or the 60th number or so would always be the one that would be the closest to 10,000.

BryceUm uh and and so yeah, I think I think to actually really get some some real answers here, we would have to do a little bit more investigation and a little bit more work. Maybe Tyler would be willing to do that for us.

ConorMy my question that I really want answered is across these two different, in fact, across the two different sequences and all the different um versions of these, although I I guess we haven't made any changes to accumulate and transform reduce, so it's really just the two different uh sequences of data. Uh Transform Reduce has always been slower than accumulate. And I'm confused as to why that is based on what we talked about earlier, that even in the non-parallel version, it's still taking advantage of some stuff.

BryceUm why is it slower than I do not think we have sufficient data to draw that conclusion? I think the only data that we have now is data that indicates that we do not have enough information to make any conclusions.

ConorUm, what have we been talking about for like the last uh to keep Bryce awake?

BryceYes. But I mean this is this is this is almost almost every time that I'm presented with performance data, um this almost always is the case. Um that that that there's some red herring in the in the in the performance data. Um the first time you benchmark something, you almost always are gonna benchmark it in in not I'm not gonna say wrong, but you're you're you're you're doing you're starting off an exploration and you may not be measuring the thing that you wanted to measure.

BryceUm the the process of of arriving at performance truth is almost always an incremental process. And once you've been doing this for a while, you start to learn to be skeptical of data. Um and the only way that you get less skeptical of the data is you go and you try new things. Like assume anytime that you're presented with performance data, assume that it's wrong until you can build a really strong case that it's correct.

BryceBecause it's so often that there is not a mistake, but but just that that just a you're measuring the wrong thing, or or you're measuring some unexpected effect, or there's some other thing that's that's influencing your experiment. Um uh and and you end up drawing the wrong conclusions because you just assumed that the numbers that you were seeing um uh were you know were actually the thing that, you know, the the the were truth.

BryceWhereas oftentimes, you know, those numbers might be the right numbers for what you were measuring, but you were measuring the wrong thing. So my my general advice to anybody doing any sort of benchmarking work is be incredibly, incredibly skeptical until there is strong, overwhelming evidence um uh uh otherwise.

ConorSo I guess I guess the question is if yeah, I mean Tyler may or may not, or someone may or may not else now go and do this as an exercise. Is there a library that you've used in the past that does all this, or is is it just hand-rolled timing stuff and oh I have little these little frameworks that I've pulled around from place to place, but the f the framework that you use is less important than than than the technique.

BryceI mean you could use any framework and and and get to the to the right result.

ConorUm I mean I'm not really necessarily that sure about Quickbench though, because it seems like uh the the charts that are generated, I've only seen two.

BryceThe charts here are not how I would generate charts.

ConorUm like how would you would you export export to Excel and then just do it there?

BryceExcel or I would use some framework or some other tool. Like when when I am producing, like a large part of my job at NVIDIA is producing performance graphs. Um and and like one one of the fundamentals of producing good performance graphs, one higher should always be better. Um and and that means like your your y axis should not be wall time. Um it should be some like there are a few, there are very few exceptions to this.

BryceAlmost always you want your y-axis to be some some metric of of throughput um because that makes higher be better. Um and yeah, and I mean the the uncertainty thing, you know, is fairly important because if if I take a measurement and there's you know 50% uncertainty in in that timing or 100% uncertainty in that timing, um, that tells me that I'm probably like trying to measure something that is too small to really be measured with the timer that I'm using.

BryceAnd so I won't be able to, you know, get any useful information out of that. Um and so frequently people end up doing that unintentionally, where they measured something that's too small. Um so yeah, I mean, like I I've written frameworks for doing this in the past. Um, you know, if I need to do something quick and dirty, I'll often just like throw it in Excel. Um you know, Google Bench is like pretty good. I have some moral objections to some of the things that it does, but it does a pretty good job.

BryceUm there's a bunch of good Python libraries for doing these sorts of plots. Um yeah, but but like at the at the end of the day, like you're when you're trying to answer a performance question, um like ideally your data should be be simple enough that you don't need a graph. Um like you know, yeah, here in this case there's the question of you know um uh of input size, sure.

BryceUm but like it it's gonna be it's gonna vary somewhat, but like it's gonna it's not gonna be that interesting once you get like beyond like small. Um uh and so like you could just like send me this data in like you know a plain text table, and I'd be able to infer the results from it if the data was presented in in a uh you know in in in the the correct way, if you will.

BryceUm so so yeah, I mean I I don't think it matters really what like graphing tool you're you're using. Like if you have if you've collected the right data, if you've collected it in the right way, I think you'll get to the right result. But but in this particular problem case, um there's gonna be um an overabundance of data. You a lot of benchmarking problems you end up with an overabundance of data where you have there's a ton of different things that you can benchmark um and different parameters, and then like it becomes unclear what things are actually um useful.

BryceAnd one of the problems here is like our discussion over the past hour, we've just identified that there is another axis. Um, and any problem that has you know two primary axes is gonna be very difficult for humans to wrap their head around. And but what I mean by that is that the assumption was that the axis here were um, you know, your uh uh independent variable was the input size, um, and also you know, what the particular algorithm was.

BryceSo input size and algorithm, and your dependent variable was time. The thing that we've just demonstrated is that there's another dependent variable, which is um the uh the input sequence itself. Um and there's actually a ton of different input sequences. There's one, there's a wide range of um rant pseudo-random input sequences. That's fine because that actually all of the different pseudo-random sequences will boil down to one data point because you'll just say we're gonna try a hundred different pseudo-random sequences with the same seed, and then we're just gonna average them.

BryceUm that'll be fine. But there's also like the special case um uh uh sequences. Those ones are probably data points that are not super interesting. Those will like tell us about corner cases, but we don't want to focus on those. Um, but that that still means that you've now got three dependent variables. Um, and uh one just in terms of visualization, three dependent variables means that you need to slice um or you need some sort of 3D visualization. And and it's very hard to like tell you know how you're gonna get useful information out of that.

BryceBut but I mean I I think the way to do it would be to um uh to try to remove one of those um uh dependent variables, and the way to do that would be uh to remove the uh uh the order or the input sequence order dependent variable, and you do that by saying, well, okay, we're gonna try you know a hundred different input sequences for each problem size, and then we will just take the average time of all of them, um, and maybe like the you know uncertainty of that um too, because that's you know a useful thing to have.

BryceAnd then you're back down to just the the dependent variables of uh which algorithm and uh the input size. But you might get into weird well the what one of the nice things I guess will be that um if you pick a large enough set of uh of these pseudo-random input sequences for each uh problem size, you're unlikely to run into too many weird artifacts. It probably needs to be like at least a hundred or so to avoid running into a bunch of weird cases.

ConorAll right. Well, we will see what our listener does with all of that information. Maybe we're gonna get a beautiful uh, you know, JavaScript rendered, you know, drop-down three-dimensional viewing of all this stuff. And uh well, I feel like this is gonna be this is our new lost reduction series where we did like four episodes where we we kept on revisiting, and now we're gonna be doing this with this problem for like two more.

BryceSo all of all of Cobb is NVIDIA's parallel algorithm uh uh primitives library, and all of the benchmarks and tests for um all of the non sort algorithms take like an order of magnitude less time to run than for the sort algorithms. Because not only do you have to do not only do you have to like think about this in terms of benchmarking, but this problem is actually far more important for unit testing your you know sorting or rearranging class um uh uh of algorithms.

BryceUm like you need to test all of these weird corner cases of sorting. Um uh and you need to have like a good comprehensive test of these different like input sequences to make sure that like you're doing the correct thing, especially in parallel. Um and and like when you think about it in parallel, um, you know, if you're if you're if you're doing this problem like in parallel, those corner cases can become a lot more important. Um uh because the the the corner cases may expose you know weird race conditions that you'd not thought of.

Ben DeaneTesting is important to sum up.

BryceYes, testing is important, is what I was saying.

ConorNot just benchmarking. Yes. Well, I'm curious to see. I mean, admittedly, this this part, this is probably gonna be a four-part. This is great.

BryceThis is good because I'm going on content though.

Ben DeaneThis is just that we could talk about algorithms and associated things for hours if not days. I mean, if I was younger me. We haven't even talked about partition copy yet.

ConorYeah. If I was younger, me listening to this, um, this would be great because I didn't have any of this. Um at some point I started at zero in terms of my knowledge about algorithms. And yeah, so did we all even the observation that like std accumulators can be used uh to implement. Like that was not something that was obvious to me for a very long time. I think probably the first time I really picked up on that is when I read the Beautiful Folds paper that's recommended um from something Graham, Paul Graham, maybe.

ConorUh I might be getting that wrong. I'll link I'll link the paper. But it's yeah, Beautiful Folds, and it just talks like Graham Hutton? Yeah, yeah, Graham Hut Graham, that sounds right. Was it Hutton or Sutton?

Ben DeaneIf it's it's Hutton, if it's what I'm thinking of.

ConorWhat did I say? Graham Paul? I wonder who that is. You said Paul Graham. Or Paul Graham.

Ben DeanePaul Graham, he's uh he's a list person. Graham Hutton is uh that's cool.

BryceAll right. Bryce, uh Bryce needs to go to sleep. Um but first, then, then some important C now related questions. Okay. Juan, are you bringing are is Ava coming with you?

Ben DeaneNo, she is not. Not this year.

BryceUm, guinea pigs need famous thing. Yeah, yeah, I understand.

ConorCould have brought the big guinea pigs. That's a COVID. You don't want them to get COVID, though.

BryceSo um my girlfriend's coming to C now, and uh, we're bringing her dog, which is gonna be a very complicated that's involved here. But uh yeah, I was hoping that other people would be bringing their partners so that she would not be stuck alone with a group of uh standard birds. Um but uh but uh okay, next question.

BryceUm which of the cuisines uh do you wish to join us for dinner for? The sushi night, the like one of the Italian nights. Um I think that that's uh yeah, it's basically like either gonna be like you have the choice of like European or sushi. So pick between those two.

Ben DeaneUm I gotta say, I'm not sushi is one of those things that I periodically try because I think, hey, you know, I should I should really develop a taste for this, and I've never developed a taste for it. Having said that, you know, the sushi restaurant doesn't only I think uh uh the the bottom is closed this year, only the top, and the top is mostly just sushi. So I'll you know I'm sure it's all probably uh uh of those two, then I would probably go for the European.

BryceYeah, but you probably want the Italian, yeah. Um so the the way that Bryce dinners are gonna work this year is um is that Bryce dinners are gonna be a four-person affair. Um it'll be me, the girlfriend, and two invited guests. So I've I I I want you will be informed which I'll let you know which life I have in mind for that.

Ben DeaneI'm on intercase.

ConorWhat is what is four? Is this a COVID thing?

BryceNo, I I've always well my my rule of thumb has always been that like the optimal group size is like um uh you know six, because if you do five, they may try to force you to a four-top table. Um and four-top tables are like usually like not actually that up that ideal. Um so usually it would be like I would take six people to dinner because then you'd be forced to get they'd be forced to give you a bigger table. They can't force you at a four-top table. Um, and like if not six, then probably three is the optimal number, um, because then you get a four-top table and you have extra space.

BryceUm uh but uh so you're choosing the third optimal number. Well, well, four four is not really like the optimal number. Yeah, but um, but if you if you're taking your partner to a tech event um uh and you're gonna go to like you want to bring some of the professional folks from the tech event to dinner, um three is not really great because then it's very intimate um uh for the tech person. Um it's like you know, you're a third wheel.

BryceUm and and you can't do more than four because then your partner is totally and hopelessly outnumbered. Um and so really the only options are either four or you don't bring uh uh other you know tech people. It's either four or two. Um so I've decided to go with four. And four also gives you the option that if somebody else has brought their partner, then you can bring that person and their partner. Yeah. Yeah, exactly.

Ben DeaneThis is reminding me that's the strategy. This is Bryce's restaurant algorithm. This is sort of reminiscent of um, is it the chicken nuggets algorithm?

ConorWhat?

Ben DeaneI hope you're still recording, Ben. Um I've got to remember so so, as you may be aware, at your favorite fast food restaurant, whatever that may be, there might be a few, um, you can buy chicken nugget meals in various sizes, right? But they are fixed sizes. And so there's a there's a problem out there of like what's the optimal way to order N chicken nuggets given the fixed sizes and the prices of them. I think that's it. It's something similar to that.

ConorOh. Something like the knapsack problem or something like that.

Ben DeaneYeah, except that might be different. People being people, they've turned it into I I think at one point there's one of these restaurants where you can order a custom number of chicken nuggets. Right? I think this is the problem, right? And so so let's say you want I want 17 chicken nuggets because I've found out that that's my optimal number of chicken nuggets. Um, how should I fulfill that order given given the standard sizes of chicken nuggets that are on offer? I think that's also related to the problem.

ConorThis is a great way to wrap things up. When I was in Sunnyvale, there was a McDonald's on Matilda Avenue, for those of you local to the valley in Sunnyvale, and I don't know how this worked, but they had a 10-pack of nuggets for $4.99, and I think they had mispriced a 20-pack for $5.

Ben DeaneRight.

ConorIt was one cent more for an additional $10. And let me tell you, I thought that was a banger of a deal, and I may have had quite a few 20-piece McNuggets while down in the valley. Nice.

BryceThere is another going back to the Bryce Dinner algorithm seating, strong preference for even number of folks, even if even part partners um aside. Um, because if you got six, um uh then you can sort of have three groups of two people talking with each other, or like, you know, like group, like larger groups talking. I think when you have odd numbers, um it it becomes more likely that somebody gets left out of a conversation, especially at like five and like seven.

BryceAnd like obviously anything above eight, and it's just chaos, just absolute chaos. Like you're never gonna be able to deal with you know paying the bill.

ConorThere's a window into Bryce as an individual that he has put this much rigor and thought into the optimal number.

BryceI mean, the other the other nice thing about four, where one person uh has their partner there with them, is that it's only a three-way split of a bill, and almost every restaurant's gonna let you do a three-way split of a bill. It's only once you get up to like a four-way split that they're gonna start having problems. Yeah. I I do I do put a lot of thought in in like there may be a spreadsheet with like you know, who's gonna be going to dinner with Bryce on which days it's C now. Lunches though, lunches are lunches are unplanned.

BryceLunches are unplanned.

ConorAlright.

BryceAlright, I gotta desperately go to sleep.

ConorThis has been awesome. This is uh this has been great.

BryceBen, we're gonna we're gonna be having you back.

Ben DeaneThat'd be great.

ConorYeah, we'll talk we'll talk about the the we didn't get to the end of the heap, but we will have you back to get to the end of the heap.

Ben DeaneI will still be reaching out to Connor with opinions. Like, yeah, it's great to to to talk to you too about uh algorithms in general, whether it be on the podcast or not.

ConorWell, I think I think the podcast is a great is a great format because then everyone else gets to I mean uh maybe not all our listeners uh care as deeply, but I'm I know Con Connor's just saying that he wants a recorded record of all of his interactions with you. Right. That's not right.

BryceI seriously I though I gotta I like do need to sleep like very badly.

ConorThanks for listening. We hope you enjoyed and have a great day.

Open the transcript file

Hosts and Guests

Guests

  • Ben Deane Guest Inferred

Credit the people on this episode

Be the first to rate this episode

More episodes

All episodes
  1. Episode 313: C++26 Reflection vs CRTP, AI, Kids & More with Zach Laine

    Nov 20, 202647 minAudio

    In this episode, Conor and Bryce chat with Zach Laine about C++26 Reflection vs CRTP, AI, kids & more. Link to Episode 313 on Website Discuss this episode, leave a comment, or ask a question (on GitHub) Socials ADSP:...

    • Apple Podcasts

    Credit the people on this episode

  2. Episode 312: The C++26 Contracts Fiasco

    Nov 13, 202652 minAudio

    In this episode, Conor and Bryce chat with Zach Laine about the C++26 contracts fiasco. Link to Episode 312 on Website Discuss this episode, leave a comment, or ask a question (on GitHub) Socials ADSP: The Podcast:...

    • Apple Podcasts

    Credit the people on this episode

  3. Episode 311: Career Changes, GPUs & More with Zach Laine

    Nov 6, 202628 minAudio

    In this episode, Conor and Bryce talk with Zach Laine about his career and the career changes he has made. Link to Episode 311 on Website Discuss this episode, leave a comment, or ask a question (on GitHub) Socials...

    • Apple Podcasts

    Credit the people on this episode

  4. Episode 310: Leaving NVIDIA for... ?

    Oct 30, 202624 minAudio

    In this episode, Conor and Bryce talk about Bryce's next chapter and where he is leaving NVIDIA for. Link to Episode 310 on Website Discuss this episode, leave a comment, or ask a question (on GitHub) Socials ADSP: The...

    • Apple Podcasts

    Credit the people on this episode

  5. Episode 309: Default Construction & CppCon 2026 with Jason Turner

    Oct 23, 202629 minAudio

    In this episode, Conor and Ben chat with Jason Turner about default construction in C++. Link to Episode 309 on Website Discuss this episode, leave a comment, or ask a question (on GitHub) Socials ADSP: The Podcast:...

    • Apple Podcasts

    Credit the people on this episode