hpr-knowledge-base/hpr_transcripts/hpr0833.txt

Episode: 833
Title: HPR0833: Ian Romanick Interview at X.Org Developer Conference (XDC) 2011
Source: https://hub.hackerpublicradio.org/ccdn.php?filename=/eps/hpr0833/hpr0833.mp3
Transcribed: 2025-10-08 03:12:49

---

This is Marcus again. This is my third interview at the XORG Developers Conference with
Ian Romanick. He works on Mesa. So this is the third interview at XORG. Is this XDS or
XDC? You know, I think people are starting to just call it ZORG-Con. ZORG-Con. OK, ZORG-Con
2011. I'm here with Romanick. Ian Romanick. Romanick.
OK, so. And I guess, so who are you? Tell us a little about yourself. I think X, what
you work on. So I'm Ian Romanick. I've been working on X-related things since I want to
say 2002. I originally got into Mesa development back when I was working at IBM and in the Linux
Technology Center there where we had sort of this back in those very hidey early days, we
had this sort of nebulous goal of of make Linux better. And so one of the big problems that
we encountered was graphics and in particular 3D on Linux was a disaster. So I had sort
of open rain to just jump in and work on stuff wherever things were bad to try to make things
better. So I spent a lot of time working on new features and trying to fix bugs and existing
drivers and just get involved and improve the state of things. And so that's what I did
there for quite a long time before moving over to Intel three years ago and where I've been
working on the various drivers for the Intel integrated graphics chipsets. Primarily still
in Mesa and now I'm the, I guess I'm the architect of our 3D driver for those parts. So what is
Mesa? I was talking about that a little bit before but it's, what is it? So technically speaking
Mesa is an implementation of an open GL-like programming interface and we have to call it
that for trademark reasons. You can't call it, you can't just call it open GL or something like that?
We can't call it open GL because it's never passed the conformance tests and no one's
paid the licensing fees. So call it Mesa. We'll call it Mesa. All right. So it's, so they
have a software programmer right against to open GL API anyway and anything that the
card doesn't hardware accelerate, it falls back to Mesa, is that how that works? Well, so, so basically
Mesa converts the things that the application calls tracks state and sends commands to the
driver which then either sends commands to the hardware or if it's trying to do something
that the hardware can't do, we'll fall back to doing rendering in software but hopefully
you never have to fall back to rendering in software because that's super, super slow. So
how much of an average application now is hardware accelerated versus using Mesa versus
well, so it's all using Mesa, right? I mean it all goes through Mesa and because the
driver lives inside of Mesa, it's part of it. So how, this part of the XM not familiar
with that at all. So Mesa is just the overall, so the OpenGL XM the hardware accelerated stuff
is inside of Mesa, right? Yeah, it's, so there's a lot of device specific code that's inside
of Mesa. And so when you build, so when you build the the DRI drivers, which DRI is the
direct rendering infrastructure, that's the interfaces that we use in X to arbitrate access
to the hardware. When you build one of these DRI drivers, you have, you build the blob of
device specific code for whatever driver you are and then all of the state and data
tracking code that's part of what we call core Mesa, which is just the device independent
parts of it, the stuff that that handles sort of the, you know, taking care of things
that are just part of the API and that aren't really specific to a piece of hardware.
How does LibDRM and DRM tie into all that? So LibDRM, so DRM, which is the direct rendering manager
is the kernel portion of this. So when you, you set up all these commands and need to submit
them off to the hardware, well, we can't just have anyone banging on the hardware that
would be all kinds of security holes. So we have to have the kernel do that for us.
So we submit buffers containing commands to the kernel. Most of the kernel modules for the
different drivers will then look at those command buffers and make sure that you're not doing
something, trying to do something hinky and then we'll send the commands off to the hardware
and then most hardware works where it, after it's completed commands or has done different
things that will generate interrupts and the kernel module will handle that part of it.
Okay. Then LibDRM is a set of wrappers on top of the raw kernel interfaces. So both,
so at least for the Intel driver and I believe most of the other drivers work like this,
both the 3D driver and the 2D driver link with LibDRM to send so they can send in commands to the kernel.
And that's especially true in when you have KMS to do to a kernel mode setting.
Okay. And that's that allows you don't have that flicker when you start our hardware
goes from text mode to QE. Well, so I mean the big advantage of KMS is that it pulls a bunch of
code out of the user mode driver so we can do a bunch of different kinds of arbitration of things
and we can do page flipping and different kinds of memory allocation.
It's one of those things that has, it's death by a thousand cuts or life by a thousand,
I'm not sure what. It has a whole bunch of little benefits that all pile up to be extremely useful.
Okay, so it's not just the text mode. Right, right. That's that's kind of the icing on the cake.
Okay. So DRM and DRI and all this is all that is within Mesa itself. Mesa is the overall.
Well, yeah, there's a lot of the DRI stuff is in Mesa. A lot of it also lives in the X server.
Part of of DRI itself is when you try to run a 3D application, it needs to arbitrate some access to the hardware
and some permissions and some buffer management that it has to coordinate with the X server.
And DRI defines that protocol between the client and the X server.
So that for example, when you want to get the buffer so that you can do 3D rendering to a window,
some communication happens with the X server. When you want to, if you're done rendering to the off screen buffer
and you want to flip to make that visible, there's some protocol that happens with the X server to make that buffer swap happen.
Okay. And a bunch of that protocol changed and got revised and changed around when compositors started becoming more common.
And we had to revamp it quite a lot and that's what ended up becoming DRI 2.
The old way that we did it was, let's do a quick hack to get some stuff working so that we can actually have 3D
and then we lived with that for five or six years and then finally said, this quick hack needs to die.
We need to do it right. And that's what DRI 2 is.
And that's, yeah, DRI 2. So when you draw, so the 3D is used for 2D stuff as well, or just 3D.
Yeah, just for 3D. So if you're trying to accelerate like a normal window on a screen, like a normal application,
does that do any kind of hardware acceleration?
Yeah, there's acceleration for render, but all that lives in the 2D driver.
Oh, okay. So there are various people who have big dreams of being able to make that live on top,
have a generic set of render acceleration that can live on top of OpenGL.
But there's some limitations of how 3D hardware works that make that challenge.
Was it the EGL stuff that looted way several years ago?
So EGL is kind of an orthogonal issue.
This was the stuff. It's had a couple names. I think the name of the most recent project trying to do that is,
I want to say glamour. I believe it's the most recent project to try to do 2D acceleration on top of OpenGL.
But for the most part, it sounds like the 2D stuff, the regular, the windows and buttons and all that stuff.
That's not really hardware accelerated?
No, it is. It just goes through a completely different channel to do it.
That's what the X-Render extension is.
So X-Render...
I thought X-Render was going away, or did I miss any misunderstanding?
There's a lot I don't understand here.
Yeah, it's going away just like X is going away.
People say, oh, it's going to go away in a couple of years and they've been saying that forever.
Okay, so it's not like Wayland's going to... Well, maybe that's a separate topic.
That's a future that's not here yet. You've still got all these?
Yeah, something like that.
Alright.
So part of my thing is I'm trying to get people to come into X and Mesa being part of that.
I mean, how can people get involved in Mesa?
I mean, is it basically learning curve like a vertical wall, like a clip they got to climb up,
or is there anything they can do to help?
So the way that most people end up getting involved,
and I think actually the way that I got involved, too, is find some particular thing
in that piece of software.
And I think this applies to getting involved with most pieces of open-source software.
Find one particular thing that it either doesn't do, or that it doesn't do right,
and start asking questions and looking at the code and trying to figure out
either why it doesn't do it right or what it would take to make it do the thing that you want.
And just start pulling at that one little thread.
And one of two things happens.
You get frustrated and go away, because it turns out that it's a hard problem or a hard project or whatever
that you don't have the time or inclination or whatever to do.
Or you start to learn about the project and get involved with it,
and gradually you pull at that thread long enough, and you completely unravel the whole sweater,
and then you realize that you have a thousand patches submitted in that project,
and you're one of the core developers.
When it feels like six months ago you didn't know anything about it.
So this is the expected, you start out, you have no idea what's going on,
you're actually in a dart.
Oh yeah.
All right.
And as far as I know, that's the way pretty much I'll go out on a limb,
and I'll say that's how every single active developer on this got involved.
There was either some piece of functionality that they wanted that was missing
or something that was there, but that didn't work right for some particular scenario,
and they just started trying to figure out why,
and asking people questions, and looking at source code, tracing through things,
and looking at how data or execution flows through the code base for that one particular microfeature
that they cared about, and eventually they just learned enough to be able to accomplish
that one small thing, and then, you know, at that point most people are hooked.
So there's still, could you even do a mace all while then?
Is there still like little threads that you don't fully understand now these days?
Oh yeah.
I mean it's a giant code base.
I mean there's, so a few weeks ago we removed support for a bunch of deprecated old drivers
and gutted out, I think, 127,000 lines of code.
And there's still half a million lines of code left.
I mean it's a big piece of software.
So yeah, there's a lot of it.
A lot of code in there I've never seen and have no clue how it works,
but you still work the main guys.
It comes to me, so.
Yeah.
I'm still, at this point, I'm, yeah, I'm one of the people that most people go to with questions.
Like, how does this, this part work?
And a lot of times I know, sometimes I don't know, but in those cases I will almost always know who to ask.
Okay.
Okay.
So then you'd be good to ask this.
So is there anything that's happened recently or will happen very shortly in mace that people will start seeing,
like six months, the next, just those come out or anything.
I think it's really cool.
127,000 lines of code is pretty cool.
Yeah.
Well, unless you happen to have one of those, you know, eight year old pieces of graphics,
if you have a G200, you may be sad.
But if you have a G200 and you're trying to run modern 3D applications on it, you're already sad.
Well, so like we presented earlier today at the conference where my team and other people in the community
are working full steam ahead towards OpenGL3.0.
So that will be exciting on that.
I think on that spot, I think I saw eight total things we're still missing.
I know those other things were underneath the cover huge, but yeah.
So there's eight, I think there's like eight primary feature groups that still remain to be done.
And a couple of those.
So there was one of the items that when we went through and figured out what it was going to take to implement it,
it was like five person months worth of work.
Wow.
Eight things seems like a solo list.
Well, so one of them was like five person months and one of the other things on that list,
I think we figured was two weeks.
So there's some variation in the size of it.
Trying to see the light at the end of the tunnel when 3.0 will be...
Yeah, I think we're going to...
You know, I'm always reluctant to say anything about dates because I'm not asking for one.
Well, because it's software.
And it's how it goes.
But yeah, we're getting close.
I think we're going to have it.
We're going to have it.
Very cool.
So if you're working on 3.0, does that mean like two or two one are done?
Or are they basically separate questions?
Yeah, I mean, we've supported 2.1 for quite some time.
So the way that Mesa numbers its releases the major number,
roughly corresponds to some GL version.
So the entire Mesa 7.X has been OpenGL 2.1.
Oh, OK.
And 6.X was OpenGL 2.0 and 5 was OpenGL 1.5, etc.
This is Mesa 8.0?
So it will be 8.0 when we have 3.0 support.
Assuming that we keep that numbering scheme,
and I don't see any particular reason to change.
OK.
So I have to ask about OpenGL 4.
Is it, what's the story with it?
Not that I'm pushing because I'm happy that you always come here.
You know, someday.
It's definitely a goal.
But there's a lot of work to be done before that.
How many of you are working on Mesa?
So right now, my team at Intel is 6.
6 people who are mostly full-time working on Mesa?
OK.
Do you know about like this, AMD, such HDI, happy working on it?
Yeah.
So AMD has, I want to say two guys,
and then they had an intern over the summer,
who had been previously involved with the Mesa project,
and they decided to give him money over the summer
to keep doing the same work.
And then who else did you ask about?
Oh, just anyone in general.
Because I know Intel.
You said that.
I just wonder if it's the overall...
So primarily corporate wise,
there's Intel and AMD.
Red Hat has a couple of people who do a bunch of stuff for Mesa,
and then VMware has a number of people.
I don't know how many people they have.
All of their people were previously the tungsten graphics people.
Oh, exactly.
They got a lot of work.
Yeah.
Zach, and then Brian Paul,
who started the Mesa project,
a project at college.
Keith, well, I'm totally blanking on names.
Jacob Bornecrands.
I'm blanking on names.
They have a bunch of people who do various amounts of open source
and closed source work related to Mesa and a bunch of their projects.
Okay.
Sorry.
It's about 10 to 15 and I'm not part of it.
Yeah, I think maybe closer to 8 or 10 total.
Cool.
So 3 out of 10 will happen.
Yeah, although it seems like it's mostly us and the AMD guys
and one of the guys at Red Hat,
who are really pushing to get to it.
We're certainly the most enthusiastic about it,
but we'll have a big popcorn party minutes done.
All right.
Very cool.
All right.
So I got the same list I was asking the other guys about.
I guess I asked if people can get involved.
I mean, something you're always looking for people to try to help.
I know it's not the easiest thing to get in.
Yeah, yeah.
I mean, you know, over the last year we've had three or four new people
come to the project who've kind of gone about the process that I described before.
They had one particular thing they were interested in and they just started pulling at that thread
and now they're trapped.
Very cool.
All right.
Hopefully we can get some hopefully this podcast to the listeners.
Well, I get some more people that come help you guys out.
That'd be cool.
Anything else you want to say?
Mention chat?
I don't know how people actually listen to my podcast,
but at least two, you know, I have a couple of friends all this.
Yeah, just if you're interested, you know,
come read the mailing list, hang out on the IRC channel.
Is that Mace?
What?
Freeno.net.
Freeno.net.
I believe I want to say it's actually hashed DRI.
DevL, I think?
DRI.
DevL?
Yeah.
I believe that's correct.
Very cool.
All right.
Thank you for your time, Ian.
The conference is now over, so I'm guessing it's your time or whatever.
Yes, it is.
All right.
Well, cool.
Thanks again.
You have been listening to Hacker Public Radio,
where Hacker Public Radio does our.
We are a community podcast network that releases shows every weekday on DevL.
Friday.
Today's show, like all our shows, was contributed by an HPR listener like yourself.
If you ever consider recording a podcast, then visit our website to find out how easy it really is.
Hacker Public Radio was founded by the digital dark pound and the infonomicom computer club.
HPR is funded by the binary revolution at BingRef.com.
All BingRef projects are crowd-responsive by linear pages.
From shared hosting to custom private clouds, go to lunarpages.com for all your hosting needs.
Unless otherwise stasis, today's show is released under a creative commons,
attribution, share alike, details or license.