In this episode, Conor and Bryce chat Henry the clock, chunk_by and more! Link to Episode 178 on Website Discuss this episode, leave a comment, or ask a question (on GitHub) Twitter ADSP: The Podcast Conor Hoekstra Bryce Adelstein Lelbach Show Notes Date Recorded: 2024-04-17 Date Released: 2024-04-19 Think Parallel: Part 1 - Scan - Bryce Lelbach C++23 std::views::chunk_by thrust::reduce thrust::inclusive_scan ADSP Episode 173: Parallel chunk_by Category Theory for Programmers by Bartosz Milewski 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.
ConorSo yes, just to recap what has just happened here. Bryce was traveling in Japan, bought a French clock in Japan from the 1900s. Well, no, technic technically the 1800s, which is the 19th century. And then traumatized the clock, which has a name and is now part part of the podcast lore. Welcome to ADSP The Podcast, episode 178, recorded on April 17th, 2024.
ConorMy name is Connor, and today with my co-host Bryce, we chat about Henri the Clock, Chunk Pai, and more.
BryceDo you know you know where I just was?
ConorIn uh conference, ACCU?
BryceNo, no, I was in Dublin. Well, okay, yes, I am at ACCU in Bristol right now, but I was just in Dublin, which is uh technically the uh the first uh ADSP road trip or the first Bryson Connor road trip. We uh you and I took a uh a bus from the 2019 Belfast C committee meeting. We took a bus from Belfast to Dublin to go visit your sister. And if I recall correctly, I basically invited myself and you were somewhat uncomfortable about your plan, uh which was just to go hang out with your sister, and then suddenly you had this price along with you.
ConorThat's a pretty pretty accurate description of what happened, folks.
BryceYeah. Yeah, but no, I I am in Bristol now. Um uh I w was in Japan for the uh Japan uh C committee meeting, and then I took a bit of vacation. I went to the beautiful art art island of Nawashima in the Japanese uh uh inland sea. Um we also spent some time in Kyoto, where I bought a uh a clock, um, a clock that is going to cause a great deal of difficulty for uh this podcast, Connor, because this is a striking clock.
BryceIt strikes every 30 minutes, um, and it does not have a silent mode. So um there are going to be times when we are recording and when the clock suddenly starts going off. And I have a small apartment. There's nowhere in the apartment that I can put the clock where it won't be audible.
ConorCan I ask a question? Yes. What prompts someone to I got actually two questions. I got two questions. I got lots of questions, but I'm only gonna ask my first question. And I'm gonna I'm gonna ask these in order, and then you can answer in whichever order you want. First question. What prompts someone to buy a clock that makes noise every 30 minutes? And question number two it never is quiet? Like do you not plan on sleeping anymore?
BryceI mean, it's not. It's a small clock, it's a French carriage clock, an 1890s French carriage clock, so it's maybe four or five inches tall and like two inches wide, um, and maybe like three inches uh deep. So it's not it's not a big clock.
ConorDoes that have does that have anything to do with how much noise it makes? Or because that doesn't answer either question.
BryceIf it was a larger clock, it would make more noise. My uh my mother just returned to the uh hotel room. It's I I guess worthy of mentioning, which is that my mother is here with me at ACCU.
ConorUm Does your mom want to be on the pod right now?
BryceUh maybe. Um but let's let's let's close out the clock, the clock's business.
ConorRight. So that's a small clock.
BryceI mean, well, you you you you want to know why would somebody buy a clock that chimes? And this is an excellent question because some v viewers, listeners, some listeners, not viewers, because this is not a TV show. Some listeners will recall that I have a dog who has made some appearances on the podcast. And uh one of the things that sets off the dog is the doorbell going off. Um another thing that sets off the dog, amusingly enough, is the chime that the lift app makes when it successfully hails you a ride.
BryceAnd this noise that the Lyft app makes, it it does not seem to respect silent notifications. It seems like they have a bug in their app where that particular noise um uh does not pay attention to your notifications volume. It just uses your like device like media volume. And so I can't really mute it. And every time the Lyft app dings, every time I get connected with the driver, it dings, and then the dog goes crazy.
BryceAnd um and so presumably the dog is not going to like Enri. Enri is the the name of the uh the clock, if you were wondering. Um and yes, I gave the clock a name. Um the the this poor clock had to had to was traumatized, uh uh being transported all over Japan and then back home. And in fact, uh is currently not in my apartment, but is with a clock doctor. Um uh because uh uh Henri has got an ailment that that requires some care.
BryceUm but it it makes a very lovely sound. And and I I was just very charmed with the clock. I was very charmed with the vintage clock store that uh uh we got the clock uh from in Kyoto. We actually we not only got that clock, we also got a small cuckoo clock, not one that makes noise, just a small German cuckoo clock that Ramona really wanted that uh got thrown in for free because we were buying the much more expensive clock.
BryceAnd then uh Ramona got a brooch watch, which is kind of exactly what it sounds like. It is a a brooch that one can pin to their clothes that is also a watch. Um and I I thought that there was gonna be greater resistance from my uh partner on getting this uh pr lovely but potentially annoying clock. Um, but no, she thought that the noise was very lovely and very soft, and she did not think that it was gonna be a problem.
BryceSo we're gonna find out when we get home and when I get on reback from the uh from the clock doctor. Um and yes, some like more uh more modern, if you will, um antique clocks um will have a silent mode. Um, but this is a uh fairly old one and not a particularly fancy one. Um, so it does not have a silent mode. Um so yeah, it at night you just you just get used to it.
BryceAnd I mean, we we we it really didn't bother us when we were sleeping in a hotel room with it. So I mean, I uh there's a disclaimer I have to put out there, which is that there's a lot of people who I have told throughout the years that, like, hey, if you're ever in New York, feel free to crash at my apartment. And I should note, listener, this this is not a blanket invitation. Um, if you have not explicitly been invited to crash in my apartment, uh, this is not for you. Um, but if you have explicitly been invited to crash at my apartment, you should now be aware that if you are sleeping at my apartment, there will be a French carriage clock that will be making noise every 30 minutes.
BryceSo it's still a great apartment. I still highly recommend you take me up on the offer. I'm almost never in the apartment, so you'd probably have the place to yourself, but uh there is also the clock.
ConorSo, yes, just to recap what has just happened here. Bryce was traveling in Japan, bought a French clock in Japan from the 1900s. Well, no, technic technically the 1800s, which is the 19th century. And then traumatized the clock, which has a name and is now part of the podcast lore. Henri the clock, and is having surgery right now in New York, back where Bryce is not, because he's in the UK.
ConorAnd stay tuned to find out if Henri makes it. And also you'll probably be hearing from Henri quite often if it's every 30 minutes, and we typically record for more than 30 minutes at a time.
BryceYes. I I I I do think he's gonna be fine. Um uh he he was he's still keeping good time, it's just his striker that's a little bit jammed right now. So this this poor clock, we we we purchased this clock, and then like the day after we purchased this clock, we had to take a train and then like two buses and then a ferry to get to this island, and then like we were in a different hotel like every night, and just in constant motion and travel, and the entire time I'm carrying this clock, which needs to be remain upright and is very fragile and can't be bumped.
BryceUm, it was a whole ordeal. A very price story. Man, so I I've been working on my ACCU talk, um, which the the premise for my talk is it it's um think parallel, and it's all it's sort of like a attempting to teach people the mindset that you and I have been cultivating for how to approach parallel programming. Um so it's meant to be really be a deep dive into particular algorithms and data structures. And originally when I pitched the talk, I was gonna like have it, I was gonna do one version of the talk and it was gonna sort of cover all the bases.
BryceUm, however, that's not how things have panned out. So now the version that I'll be presenting at ECCU will be think parallel part one scan. Um and so instead of a talk that touches on a bunch of different topics about parallel programming, it will top it will touch on one topic. It will be 90 minutes of scan, where we talk about how to implement scan, how we can use scan, um, all sorts of things about scan, and I am very excited for this talk.
BryceAnd if you are a listener, some of the things that we have talked about in this podcast in the last year will certainly come up during this talk. Um, and uh right in the like three or four days before I got back from Japan like a Sunday, and then my mom got in on a Tuesday, and then on Thursday we flew out to to Ireland. And so I had like between that Sunday and that Thursday to um to make my talk.
BryceI'd already given an earlier version of the talk, but um I I needed to like write this algorithm, this this new chunk buy algorithm that you and I've talked about a little bit. Um and uh I needed to like write it and then like make slides for it. Um and then like also like we had like dinner plans and like I had like no free time. I had to unpack and repack. Um, but uh I got some good news, which is that it it did require me staying up till 2 a.m.
BryceBut uh at 2 a.m. I I cracked it and uh I had a little silent party because at this point my mother was asleep on the couch, my dog was asleep on the couch, my girlfriend was asleep in the bed, and uh everybody else thought I was a little bit crazy that I'd been working on this all day and all night, and had been just a distraught during dinner that I could not crack this algorithm. I was there at my favorite sushi place, my mother and my girlfriend having a good time and chatting about, you know, life and non-work things as one should at a nice sushi place.
BryceAnd I'm sitting there, just like in the zone with a piece of paper and a little notepad, sketching out math and stuff, and uh and they're just like politely letting me be. But I cracked it. I cracked it. And uh it, Connor, this this uh this algorithm, when you see it, it is gonna blow your mind. It's beautiful. It is beautiful. And I benchmarked it too, and it's fast. It's like uh, I don't know, it's like eight or ten times faster than the uh than the multi-pass version.
BryceUm yeah, so uh it it's it's it's great. It's great. I haven't been this excited since uh since I cracked that rolling reduce algorithm uh last year. And that rolling reduce algorithm will probably be a whole other, I'll probably do a think parallel like reductions talk. Um although that rolling reduce is actually really kind of a scan um because it's implemented in terms of two scans. So this whole think parallel talk series might just all be scans. It's all just scans.
BryceThat's all the that's all the really is. Anyways, I'm uh I'm excited about it. I um I I got into the conference today and I did not go to the conference at all because uh I needed to like actually finish making the slides. And uh I've got most of it together now, um, and I'm pretty happy with the result.
ConorWell, first of all, congratulations. Second of all, you said that we've mentioned chunk by a couple times. We had a whole three-part series. Chunk by part one, two, and three.
BryceLook, man, it it's all my my my life for the past six weeks is all a blur. Like I have been, I was I was in Japan, then I was in Ireland, now I'm here. Uh lot of like at some point I was in New York, I was writing a lot of code, doing a lot of things. It's been chaos. Um, but yes, I I do recall now that that there was a there was a three-part series. Um and uh the the algorithm that I poorly describe, I think, i in that recording is the algorithm that I have now successfully implemented.
BryceUm and one of the So the thing that made it trickier than I had initially anticipated um is uh the boundary conditions. Um uh which is that yes, you can implement this with a scan, but you sort of have to be clever to um handle the uh the first element and the last element, but also then if you're tiling it up and doing it in a single pass, you have to do some trickiness to handle um the beginning and the end of each tile.
BryceUm and uh at one point I thought that there might there's this thing in Cub called block discontinuity, which I think is how you would do this if you were doing it in Cub, but I just wanted to write this straight in standard C. All these algorithms, everything in my talk is implemented purely in standard C parallel algorithms. Um uh but uh but yeah, I was eventually able to sort of crack that aspect of it. And I'm not sure, did we talk when we talked about chunk buy?
BryceUm, I know I posited a solution for how we could how it could be done with a single scan pass. But how far along did we go on that solution?
ConorI don't recall. I think we got hung up on the fact that we were digging we were going down the rabbit hole of looking at the reduce by key thrust algorithms and whether or not those were implemented with I think we determined that it was implemented with uh inclusive scan by key. And then we were talking about whether you could implement it as a reduce. I think that's where we kind of left stuff.
BryceRight, right. Well, it turns out you can do chunk by with a scan. Um and the the key intuition that one must realize to get to the chunk by with the scan is that first you must realize that you could implement chunk by with multiple scans. Like if if if I could do like three scans, um three very simple scans, I could do chunk by. Um it might be inefficient, but three very simple scans and I could do it.
BryceUm uh, you know, like I could I could do scan, I could do a scan to figure out um uh you know how many groups there are gonna be and what the indexes of those groups are are gonna be, and then I could do a scan to figure out the you know the start and the end of each one of those groups, um, or I could do one scan to figure out the start and one scan to figure out the end. And all of those would be fairly simple scans. And then the only tricky part that you're presented with is how do I write a single scan that does all three of those scans at the same time.
BryceUm and that also can also does the like the flag setting part too. Um and that is where the challenging part is, because then you have to define your own new data type, and then you have to define a an operator on that data type, uh, you know, a reduction operator on that data type that does the right thing. Um and one of the places where I got hung up a good bit um is on the identity operation um that I defined my operate my data type and then I defined my plus operator on that data type.
BryceUm and I defined it in a way that it worked. However, um to define a data type that's gonna work correctly with a parallel scan, you have to support a identity operation that at the very least can um be a left identity. So you you need some element that you can take that element and do that element plus something, and you always get back something.
BryceAnd the reason that you need that is because you're gonna have some temporaries that each, you know, once you split up your data, you're gonna have some temporaries that uh that each thread is going to accumulate stuff into and then communicate to the other threads. And in the in the case of something like a decoupled look back scan, you're not gonna only have those temporaries within a thread, but then you're gonna have temporaries that um will get created and stored and communicated across different threads and will get added up in interesting ways to implement this decoupled look back mechanism, which is a whole separate podcast episode we should talk about and will also be covered in my talk.
BryceUm But um uh for all of that to work, you need to be able to have a nothing object that you can add to something and get back that same something. Um and the original version of my code didn't work correctly um uh once I comported it to the single pass version because I had not thought about this question of like an identity element. And I did not have um the operation defined in a way where I could uh I could add a nothing thing to it and still get back, you know, the the identity.
BryceUm and so then I had to sit down, and that that was sort of the insight that I had the night that I was uh at the sushi place, um, was that I needed to have some identity operation. Um and then I had one other insight, which was uh just sort of an insight about how I could reset a particular count in a scan. So sometimes in a scan, you're counting things, and at some point you might want to reset them. Like, you know, like it's very simple to do like a scan that just like counts all of the things, right?
BryceThat's just like a scan with like, you know, uh a plus operation. Um uh uh that's just sort of our our typical scan. Um uh well, uh maybe that's not what not a way to put it. Like if if if um if you do a scan where um you have uh uh an integr an integer plus and where every element is uh transformed into uh you know one, uh then that'll give you a count.
BryceUm sorry, where was I? Uh oh yes, yes. So um uh sometimes you want to do something with a scan where you have some reset condition. Um and I was having difficulty getting the reset condition correct for chunk by, like when when do I reset this count for the next chunk? Um but then at some point I realized that um I had this other piece of information that I was tracking in my interval type.
BryceAnd that piece of information was um what is the index, the current index of like in the output? Like what number group am I on? Um you follow, like you like one of the things that you need to keep track of is how many um chunks have I created thus far?
ConorYeah, makes sense.
BryceYeah. And and so, okay, and so then I realize, aha. If I have the index of whatever chunk I'm currently on, then I can tell when I need to reset easily, because all I have to do is say, hmm, has has the inde like if I have two elements L and R, if the index of L is different than the index of R, then I'm at a place where I need to reset. Um with the with the caveat being that there's a special case if uh if one of them has an index of zero because you may be accumulating something on the right that's like a local thing.
BryceBut um uh that's the basic idea that got me to the correct um solution as opposed to the thing that I was doing prior to that, which was um far more complicated and stateful and just simply didn't work.
ConorWell, that sounds like it's gonna be a super interesting talk.
BryceI mean, the one of the things that I thought about I I I fear that you may be the only person that'll understand it.
ConorNo, no, I I don't think that's true at all. Definitely we should one of the people that I thought that I I think it's he's come up on the podcast before for sure. We just haven't reached out to him. We should have on Bartaj Meluski because we are struggling through the terminology of like identity value and stuff, but all of this stuff has names. Like we're referring to basically as a monoid. And then I even looked it up. I was like, what is a monoid that requires commutativity? And uh Chat GPT or GPT-4 to be precise, said that those are referred to as commutative monoids or abelian monoids.
ConorSo basically what std reduce requires is an abelian monoid.
BryceAnd I actually think that in this particular case, that I actually don't need it to be commutative. That in particular it's an I I I'm I only need an I because scan is non-commutative and because it guarantees that things are going to be done from left to right, um I only actually need it to be like a left identity. That this identity element only has to work when it's on the left-hand side of the operation.
ConorInteresting.
BryceAnd what one one thing to watch for in in my talk, and in if you I have the code on GitHub, if you go and look in the code in GitHub, one thing to watch for is whenever there is um an addition or you know, a scan operation. Um in my my code examples, I just use addition for simplicity, but whenever there's a a scan operation, the the choice of which thing is on the left and which thing is on the right is very, very purposeful. And in some places it seems unintuitive.
BryceLike there's a couple places where you might look at the code and you might be like, well, why is he using um assignment operator in plus instead of using plus equals? And the reason is that using plus equals there would put the would do things in the wrong order. It would put the thing that should be on the on the left on the right. Um and I'm fairly certain that I got it correct in all the places. And I think if I didn't get it correct, um the code would not work.
BryceUh because the data type that I'm using for chunk by is not at all commutative, I believe.
ConorWell, thanks to GPT-4 once again. I asked, what do you call a monoid that only requires the identity element times your type equals that same type? And it says that this is typically referred to as a left identity monoid. Which is GPT-4 making that up? Or does it actually the case? I'm sure there's someone that's a category theory expert out there. But uh which is why we just bring on Bartage, because probably to the category theory informed person.
ConorAnd that's the thing, is I worked my way through his book. And so when I hear you saying, like, oh, it's a type, but I didn't I didn't uh define this like nothing, you know, all of this stuff is actually already like there's a you know, category theor theory for it, and all this stuff has names, and we should do our best to try and know those names if we're gonna be it it it pains me to say it.
BryceUm and it pains me especially to say it with my mother in earshot. But um this is one of those rare times when my college education actually taught me something useful. For for those who are unaware, I dropped out of college the first time, but then the second time I uh was working at LSU and was a student part-time, and then eventually got a degree there. Um and my degree was in applied math. Um and I actually think I would have been stuck and given up on this problem if not for the realization that um the issue that I was having was that I I did not have a sufficient uh uh I could that I could not define an identity element for this uh uh for this operation, for this group.
BryceUm and uh that is certainly something that uh uh I think came from my actual math schooling. Um so yeah, that there is something that I think I learned in school. And and as soon as I had that realization, I I did something that like every now and then, every couple of months or so, there's a time where there I come across something in work that requires me to like go and do math. Um and that's what I was doing at the sushi restaurant.
BryceWas I was like, once I had that realization, I was like, all right, I'm gonna sit down and I'm going to approach this not as a programming problem, but as sort of a mathematics problem. And I'm gonna see from first principles in math, um, can I define this these operations um in a very abstract um uh pseudocody fashion? And then can I can I come up with a proof that this works? Um and I I do somewhere in my apartment in New York have the uh sushi uh restaurant paper where I I I did the proof.
ConorWell, there you have it, folks. Lots to look forward to. If you're at ACCU, well, when are you is your talk before or after Friday?
BryceIt's gonna be Friday. It's gonna be Friday.
ConorIs it Friday in the afternoon? Uh six hours ahead, Bristol, right?
BryceSo Yes. Yes. This podcast will air before the talk.
ConorOh, perfect. Uh so you know, go to Bryce's talk if you happen to be at ACCU. If not, link will be in the show notes for when it goes online.
BryceWe'll get an update on on the I do want to circle back to my my fear that my talks are not understood, which um I I fear that I lose people when I give talks. Um or I I know that I lose I lose people. Like, and it's funny because on the one hand, I think one of the things that makes me a good speaker is that I I try not to have too much assumed knowledge and I try to do things very from first principles. And I think that's one of the things that's made me successful throughout my my time in tech, I think has been um my uh my desire to like not uh assume too much knowledge when presenting things.
BryceUm but on the other hand, I uh I can feel and see when I'm when I'm losing people um and when I'm when I'm moving too fast. And when things that seem like very a very obvious set of conclusions to me um does not necessarily track for for everybody. Um and uh I do I do sometimes worry that uh that I lose people um and that uh and that I'm not sufficiently communicating my ideas.
ConorYeah. I wouldn't be too worried about it. As long as it's the people who don't think they're losing folks. It's like, you know, uh pe crazy people aren't worried about being crazy. Or no, wait. Yeah, yeah. Yeah that's exactly it. It's like if you think you're crazy, you're probably not crazy, meaning that at least it's on your radar. I know all the time that I'm losing folks when I start talking about, you know, the hierarchy of combinators and combinatory logic. And a bunch of folks point it out in the comments, but like one of my favorite comments on one of these videos was like, you know, uh not useful in my day job and was a bit confusing, but uh I appreciate the guy's enthusiasm.
ConorLike he's clearly excited. And uh, you know, as long as I feel like they are able to pick up something, even if they end up getting lost at some point. I mean, I there's tons of talks that I've watched, not because necessarily the presenter was doing a bad job, but the requisite knowledge that you needed in order to be able to keep up, I just didn't have at the time. And then you go back back, you watch their talk later, and potentially it hits, like in instead of only understanding 10%, you understand 90%. So it's uh you just hopeful, hopefully you put, you know, at least intermediate or expert on the label, and then people won't be disappointed.
BryceUh yeah, I pretty much always do.
ConorYeah. Yeah.
BryceBut it's like it's it's also like one of those things where it's like I think I barely have an understanding of how and why these things work. It's not so bad for chunk buy, actually. For chunk buy, um, I think it's pretty straightforward to uh to understand. For the rolling reduction one, there's a component of the rolling reduction thing that um I kind of have a mental model for how it works, but I maybe I don't. I cannot explain how it works. It works.
BryceOnce I see how it works, it's sort of like, oh yeah, that makes sense. But I cannot explain to you like the the the how one gets from the problem to this is how you should write the solution.
ConorYeah, it's too bad. That was one of the episodes we lost. Um which was quite entertaining, but anyways, maybe we'll revisit it.
BryceWait, we lost that? Oh yeah.
ConorYeah.
BryceUh we'll definitely revisit it at some point. Because at some point I'm gonna give a talk about rolling reduce, and to give that talk, I'm gonna have to come up with an explanation. Anyways, uh uh so so so I I mean if if this is gonna go in the episode on Friday, I guess we should mention that if you want to meet Bryce's mom, uh she'll be at his talk on Friday, and uh you could in theory find her and me uh after my talk. So yeah, go to Bryce's talk, get to meet Bryce's mom.
ConorAll right. I will I will splice this back into the first episode uh and and they'll have to you'll know that they'll have to be like a I don't know what you call them, an avid listener, because there's probably a short window of time when when the uh the podcast goes live and then your talk starts.
BryceSo I will like I will be genuinely impressed if somebody comes up to me after the talk and mentions that they have listened to this episode.
ConorYeah, I don't even think it'll it'll happen. I'll say this.
BryceThere'll be a prize. I I'll say this. If you do, there's a prize. There is a prize.
ConorAll right, you heard it here first, folks. And when what time actually is your talk at?
BryceUh 11 a.m.
ConorOkay, no. This uh London. I told you, bro. I told you. This the time difference is five hours. The time the podcast gets released at nine a.m.
Bryceuh. Could you make an exception?
ConorUh sure, why not? I will I will release it at 9 a.m. local time for ACCU, which will be, what is that, five or six hours earlier? Well, am I gonna wake up?
BryceYou know what? I sure uh you could probably schedule it.
ConorI probably could, but yeah, I'll look into it. I'll make it happen, folks. If I have to wake if I have to wake up at you know 3 a.m., uh it's not it's not the most difficult thing I've had to do for ADSP. Be sure to check these show notes either in your podcast app or at adspthepodcast.com for links to anything we mentioned in today's episode, as well as a link to a GitHub discussion where you can leave thoughts, comments, and questions. Thanks for listening. We hope you enjoyed and have a great day.
BryceLow quality, high quality, that is the tagline of our podcast.
ConorThat's not the tagline. Our tagline is chaos with sprinkles of information.
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:...
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:...
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...
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...
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:...