Bestand wählen
Merken

Wine Development Updates, Performance and the D3D9 State Tracker

Zitierlink des Filmsegments
Embed Code

Automatisierte Medienanalyse

Beta
Erkannte Entitäten
Sprachtranskript
OK so let's get started hello and welcome to my talk about line what I what I'm going to talk about them Mr. from the singer uh and I'm from called the 1st of the day and proficient in here on behalf of divine project I'm what I'm going to talk about today is the uh some updates what changed in mind in terms of graphics over the last year I and I am also about what a month to take the opportunity to address this gathering and volatile opinion of their territory in 9 states TREC that has gone into me say in a few months ago I've just just a quick question is my voice understandable like not enough no OK so let's just go ahead so we added the outline progress reports and made you're ready the thing that directly the 9 state tracker in Mesa is on a bad idea and I want to underline now our why we think that this are numbers and then later on gold % some ideas how we can improve their very very have to improve how we can improve their open-shell interface between line and research but 1st for divine updates the big changes that happened in there in the past month so that that the finally are able to handle focus loss of screen there the applications which means that when you're running a game to press on most of the time it's inside top part than their Windows gamble minimize finders stories screen resolution you can use a desktop and then you are done checking your e-mail you click on the game again and it opens up and depending on how well the game is written in the game realized to focus loss and false the game in the meantime over their rights planning to do I am the this this works on on OSX it also works on uh on KDE uh if the W M and for that to added it to the slides and also works on accepts the unfortunately there window minimization does not yet properly work on the test city and any of unknown compass and any and every of work from that the city which means Marco from my it make their stalled not the from shell and I forgot the name of the window manager that the that she notes for the non show this using but the main problem here is that on Windows games usually created window with other minimize button so we do not set out there minimize function hint which is intentional because of the dominant minimize button on the window when the game does not want 1 but the games that advance to minimize itself the wind 32 commands the callers x iconify window when that happens and the major refuses to minimize the window and puts it into a broken state rich for most purposes means that the game is no crash everything your accelerates inaccessible because of the frozen window and you get 3 subjects ah I sense but reports Ludivina managers and of I am I am discussing the proper fix for that what is somewhat related to focus handling and resolution restore it is that we need a solution to restore at this very Randy became crashes you you know because it just really crashes or because it just costs exit and thus of cleanly restore Demeter down there very he set up having gets through suggestion about that later you know start that's who the properly working on that we can on the progress of that is low but it's 6 uh but if we are making progress on under you recently finished the dot the 10 texture something supports the main big thing that is still missing our resource handling for example the can sample from but because we cannot render to buffers and we do not supports the 3 interpretation that there would be 10 edits but the reason why our roof is on fire now because of the we 10 use that Microsoft Office uses it it doesn't use it directly uses a different library for direct to the rich Microsoft implemented on top of the 30 10 and you have to do the same because the barrier to the interface exposes its that you can the foundation so I will present final he so this shows this
is this is pretty important right now and it should it should be finished within an hour the help the year this part of our diet can rightly another king and switching to Open shape core context rather than running everything that legacy context this extension on that I mean mostly needed on record records which does not expose some new functionality on legacy contexts and I hope that this is making some things easier for the driver and I told that these are has some some expensive for effects and we use school contexts request also itself so I hope that this in this this remove some performance obstacles that are currently in our rate the Our sending patches for that I think we may be done with that in a month will do and the main the main problem the facing our direct thrall it's which are in a common cold but of everything and you have to replace them with proper buttocks buffer-based roles that draw takes a triangular blow things with effects of of objects and so on and direct draw collecting emulation currently uses the test which demand which is also removed from quot context tha I think it was about to use 1 and a half years ago and I published opinion you of my command screen patches what they essentially do is that they move their direct but the 3 D callers to where different richer something windows does that improves game performance by using 1 was you you walk it also improves game performance because it allows us to want do there thread synchronization of the game courses remarkable its ourselves by just doing synchronous rights to all and a not free ring buffer rather than colleges Russian-Jewish and everything the status of that is there so it wasn't the published that this work is block on the director the 10 resource changes because it's a lot easier to switch to about the 10 style resource management when we do not have to take care of that synchronization at the same time so all the TA on that is unknown it's once the 10 standard move ahead with this in terms of performance striking over the years and I am running pretty much like a test of various games and benchmarks and on since I'm using the 4 on excessive but that you have to read the results in a different direction because it starts at the new ones and in fact back to the old ones so these 2 things here regressions 1 of them I was able to isolate the other 1 a tries to downgrade me signed line and it got even worse performance than a head pose actin regression so I don't know what the what the other down is why do a show this ages old phonemic 2000 although that's because that 1 has to changes rather than every other benchmark that just shows this 1 benchmark that if 1 regression so what this graph is just more interesting to look at on yet then this line patches posters regression is a bit of a tricky thing we use this to make our games that you state that he and open-shell at the same time work so all games and initialize Open shared and if you read the video memory size about redhead that initializes another open-shell context changes depicts the format of the window and boom the original jails of the spoken but they also that batch was looking so that the solution that involves creating a child window was very probable of material and by the and just switching between a child window on represent but it's he got into an argument of Alexander have been through the process on obvious that this Everton but still solution and I don't know and then they're going then you have in the it will be able to move to move on with that good but it as you can see except for this thing performance is mostly static and it has been steadily for for the past 2 years Rich made it is is a bad thing because performance and that it's not a good thing because find has seen increased development support from new things has suffered for new things and most of them have been developed performance regressions I'm always testing this performance on their radian 308 gallium driver so it looks pretty much the same 1 trick word about our development environments and stealth and conspiracy theories that you only care about makura so we only care about the video we only care about this so that well you know
what they can be so all Andries doing his development on our 600 xi which is currently the only driver that actually has a solid testsuite I am since I'm using this make clapped the appear using then binary driver and to true over could call the 1st to be given and the of was to keep an eye on Mac OS sense when that stops working in the Our of ponentials notice that so we have to keep track of that anything prior to direct extend cards is pretty much withdraw thing we do that but reports for it when something breaks and you try to fix it as
soon as we can but no no developer is using any Derek x 9 or the economy 1 we try to keep it working sensors this is especially troubling thing is that Internet Explore uses direct role so when you break the old constant Internet explore breaks and this is used in a lot of officers old card so we get reports and is given something goes wrong but no real development happens on that and 4 in the GP use I'm sorry I don't know on the have pretty few but reports about that I'm no developer is using it which you that you but it would see the mean that everything is working just fine or that nothing is working in everybody gave up so this is anyone here using blind wine on an integer view knowing it as do you have any problems events yes but you had to 0 it had and the self if something is terribly broken please tell me because I have an internship you in that elispot I'm only testing something then there is a part of what something is broken and I can note I do not have like you you that I do not have performance comparison put into the drive of from the windows because it cannot exist introduce GPU on Windows due to their biosimulation style far at the added on this laptop and putting Windows wiII on this level that is pretty much impossible and even and even if people who managed to prove that can exist in which if you because the refuses to note so 0 sorry no ordering the about intention to use so let's get to the main part of the British here this may be may be a bit of a controversial topic either directly 9 States trek up on I like that dates been written so thank you to X the media and all the others who worked on it because it's so I think it's a pretty useful to work with the backing the performance problems and that understanding them but to really do not see it as a long-term solution and the main reason for that is that it is a gigantic cult application for just essentially 1 corner cases and let me explain what I mean with this 1 chronic is so we have like every generations of GPU is we have multiple vendors of TP use we have currently usable 3 client API from direct drawl tool directly the 9 which we support with 1 cold face and the main main use the main operating system targets are and always can but 93 d a strange as it may sound also has a is used on Windows as that is mostly by forward game so when when games shapes and all an ancient direct role in that doesn't work anymore sometimes they use of EDI in our minds direct draw library and what and my new would need to make it work on our on you have in the systems so we we have a use case for this on Windows even though it's right limit that's what within the users and everything here is supported with 1 cold facing line and but that the guy 9 state record and practice is only usable on the then radio on the directives 10 and 11 class radiant she views it is in the reusable on NVIDIA GPU says no but if people care about performance on immediate you use they are more likely to just use the binary driver instead of using the DC United States trek up at that's what I think and the other Our called base that supports order operating systems order client libraries is about 80 thousand lines of cold including data from dense with direct or that you need about 59 executing the tests and of that the 9 state trick is about 20 thousand lines of codes and it's just bullets extend the 9 on 1 of those cases and the it will never be able to replace that 80 thousand that course the for ever have to maintain both of them and this is this is our main concern at that the it's a fixed the capsule then you probably need a scalpel to fix the fake so that the performance also that the it is integrated in vying is problematic are the main problem here is that gallium is said internal interface for me itself you do not want x alive external softer the point directly because of the the the change if and on course on the other side online the interface between or that the united front end and how the interface of x 11 is an internal interface but we do not want to publish any any internal stop you because 1 to be able to to the changes in the face and that makes into creating the service line in that may prove difficult using the parity 9 in the interface is directly as the interface between the 2 libraries is not possible because of their the 9 interface process window handles around and hot cold some assumptions about the 32 to API that varies innovates candid on needs a bunch of cortex that excess fine internal data and our as I'm going to show next is that the 9 state record does not sold the actual problem interesting separate but doesn't really solve the problem so the do that I have some benchmarks just for the sake of completion this is so might my test machine is about 4 years old by now so they're the radio and chief you think that 1 is already there 11 capable as well as their the cheapest vote it effect it has the same computable twinship use in it which makes the comparison between the data read and they're tuples
whatever this thing is going of which makes the which makes it easy to compare the numbers between the region and their the chief was GPU their windows benchmarks with them on Windows 7 are the Linux benchmarks are online Chen for installation I put this if you try but rather than believe my benchmarks reproduce them yourself and this report communicating different result because benchmarking those things tricky and it is very easy to run into we have some set the problem that gives you bad numbers and then you draw conclusion based on a bike your benchmarking so all of these if you do not do not believe my numbers pride and testing yourself and then you end up something different please contact me so when you test things on Linux of at 1st there the performance improvement due to mind looks pretty impressive it's about the like inhabit to is about double the performance finally using this sort game it's about 10 years old by now but mainly because it's the electrode of our test case for the source and hand and the engine in the game has constantly been been upgraded and always so it does not have any obvious facts you can say yeah this game you know just because it looks to render target or this game is slow just because of this this or that arm but still it it makes very in that it it uses pretty much every directly 9 feature all of them but uses them in a good way I with my comments green information I still do not the gallium line which makes this even more impressive but so they're like they command stream and 9 the native export so looking into the details especially specialized on the command screen is slower than the 9 is that main line and my comments you must evoked by the same bottleneck which is there holds that the mind is pretty cold and they're open geolibrary it's just that by doing moving is 2 different Brett I've move like a 20 % game logic into some areas so again and it the performance improvement but there the essential for tonight it is still the same so there is a problem that that's when he compares the windows still beats portal solution by like effect of 3 the and the problem is I the that the problem for me so it's not so much data the problem is this that this part is what needs to be fixed in there that that needs fixing an not so much a a whole whole very 389 performs on hope of other 39 games perform because if you it improves the 39 yeah the only improve 39 you will never improve native native in those games like things and then through open shell so all yet it did this is a personally I I see the problem trend we get but to show that there are opened their open interface is like it from this you could computed open-shell is just a terrible things for performance since all of all of those except for guide for 9 which is like faster Everything talks to me of open shared but if you look at some the dead and many of log you see there actually doing open open-shell and doing that the 9 on top of open shared bayes possible adequate performance like the the naked and what is essentially I binary Ordos this set the periods should us CPU in so I deliberately renders on a low resolution but no until lensing so this is just CPU overhead and talk about their graphics card performance next but you can see it's possible to implement open shared with low CPU overhead it is possible to implement their 3 D on top of of open shared this flurries of the CPU performance related so then that this is essentially thing that we do not need a diff as 2nd a different interface between our 2 projects tool to get good performance we don't have to reinvent the wheel that is this is the core of my argument are just a rule to look across the and some extra testing it looks like a big grand in a chip you were limited situation and it's funny that are their performance of both the radium cotton comunicado pretty much the same on on Windows it's like a 1 % difference it's the same on like images in security medicated source of 600 frames per 2nd those living 605 frames per 2nd but this is just in a happy coincidence that 1st of all that it might benchmarking wrong but I couldn't find any bike India and I think that those constants that those cotside essentialism generation so it's I somewhat expected that they have similar performance on there the unfortunate thing is I could not test their have left you in HEP you limited set up on research because it just insisted on turning on these things so I was most of the time stuck at 60 frames per 2nd this guy 90 recent but did not happen so I got about 1 3rd of having this performance even in a cheap you limited set up and sometimes it dropped below 60 frames per 2nd and just by guessing from there into looking in a benchmark run the performance drops below 60 frames I think there of both us there the Linux native version and find with the comment stream land at also blasted plus a minus 100 frames so I think of the performance of 10 is pretty similar and here we see a difference between the game running on with their game running on the Windows drive are then the given those drive and became running on the and media Linux trial and that's something there then did have people maybe maybe no maybe model that differences but I suspect there is 7 thousand at some inefficiencies venue translate directly 9 traders to the cell and then combined into a hot issue us an I the I have 1 what particular suggestion later that mediate the tool improved this but not at present entirely but I also look at other games you for example is a civilization 5 that's a game that I order that I could not get this to run it you you limited fashion it's sort of a CQ limited I suspect because of the heavy game logic that's I think written written in Python or something and again we see on brought proper performance is possible than needed level demonstrates that and I on online you know when I it so the performances with when the topic mainstream is it's this is pretty much exactly the same as the Linux Linux port of the game also runs at the same performance which is again 1 about 1 3rd of within those of what's 9 is passed but unfortunate this result is not really useable because it render visually different a different way it may be a feature in the sense that it stopped rendering some key there's consumed of it may be a factor I did not notice on at that 1st look the same rendering difference on Windows but I forgot the take and then take a screenshot so unfortunate don't really know if the lower
quantities of backward feature 5 years of age does not to make those unfortunates not make those values comparible just edited for the sake of like what completion and I am sure like if this is just 2 games but the result of match my genera observations of other games actually and even have to talk about that our gallium lying in an hour after my talk and I've seen that he has in some games for might be different numbers so this is the 0 sum games I need to investigate but I still believe in the generator general conclusion that's the problem and the performance problem is not the interface between our 2 projects I mean what 1 thing that has been in the freshly me especially concerning mental I they use mental interface is saw drawl overheads and gender over an antenna overhead in our in open chair land there is the inner got mean and no overhead is source of what the 9 Develop was a tribute to the improved performance and I agree with them on that part so gallium 9 removes a lot of overhead that you be happened around provided for the GLM than the self but it's open shared is actually pretty reasonable interface for for that that can be implemented with a reasonably overhead and I believe it's not necessary to reinvent the wheel to know what role overhead and or a soul no drawl overhead is not as a it's it's not the haughty grade so all let me show you what I mean is that like it on I have a fitness in protest programming by the 9 and open shared just costs 6 thousand tiny trolls the frame just the thousand cubes 1 side of 2 Cuban draw color and then flips and with that on their HEI cards I get on by 50 evolved to a million 300 thousand brokers the 2nd on Windows the windows of open changes is considerably faster but this has a lot more overhead so that's about 2 million slightly under 2 million blog posts and no I believe the reason why open shell is faster and you'll be losses see this on the invidious side is that very the nineties an interface therefore does a lot of babysitting so if you insist on using the output of a blending of the destination benefactive at the frame of smart for channel that's just fine of you may get a recent reproducible results is essentially comes down to using a fixed what effect or if you insist on 21 at that festival that block by patched all no problem about the given something reasonable for you and this this this babysitting I think helps a lot when you were learning and learning how to use the interface you don't run into those problems they see a black screen but there are those checks those checks make draw make cost every they're more expensive and slow down applications for the developer knows what they're doing and you see the same relation between dark you know open shared on then we decide on Windows what is different in media properties that it just beats the hell of of during this trial in terms of profit performance and the I do mean not know why they why their indeed a driver in this situation so much faster than the but in the Windows Driver I suspect that because Windows ring windows at some abstraction layers Goldman shared have Microsoft provides the opportunity to look still a library and that 1 thing causes the installable clients for driver and know that but that's just there's suspicion I cannot look inside Windows with yes I know him I don't I did say I savored their composite thing on our systems yep yeah so and now what's gives you a bit of caution here in terms of edge imputing fast performance allude overhead if you compare their it they can result in red and then results in green you see there on the on the 3 D & Media meets there it came the carts quite a bit but the end result in the game was the same so this friends apparently does not matter all that much because they gave it to all I had met too much tension excuse would be a good benchmark and as everybody knows due next is effective pretty terrible benchmark well let me correct due next years is a good benchmark for so so powerful and but it's not a benchmark for Generale ventures performance but there are some as I could see in my performance you know what time there is a correlation between drawl ahead and actually game performance of if you go back to my opinion of 2000 performance test but you see it's dropping down all after mind 1 that 7 of 13 and here is yeah the results from which all over his bench drops down very pretty at the same time for the same reason so there is an overhead but don't make draws are out there there is a correlation between the cost of localizing game performance but stone to me I draw overhead you knew God it's not it something to keep an eye on and use stroller here it is so pretty high but I don't know it's if they're the troll if if they're drawcard benchmark it's like here so I don't use in his people it's not necessarily causing problems what yours seen benchmark is what happened in this area of line on this changes and that's that something else was afraid of it's just slowly slowly slowly deteriorating it can't get it can really be attributed to a single change like this it was 1 thing to change because the changes probably broken but yeah I can't really say what went wrong and no this is really annoying because it's really really hard to to get the performance back when it's deteriorated 1 patch of than other because for every pet you can yeah we need that for additional feature yeah we need that for correctness yeah we need that because it makes 1 game faster difficult difficult situation so what can we do to improve things from mother I caught a specialist but it's more as suggestions to to start a conversation 1 the cells although overdose things also have once been shortly mentioned on their meaning is so they're not really new 1 constant problem for us AIG elicited trader compile times I'll been those games expect that they can feed trade us during rendering and creating . 2 9 traders is really fast link Forest on notice said and we have some games for example a mass effect through that kind that kind of
1st frame and then dualisms starts compiling a shade as the 1st frame takes 10 seconds and that screws up internal timing in the game because like you have 1 10th of a frame per 2nd for the 1st frame and then it runs of 1 in 20 frames per 2nd and then order timing is screwed up but the line has some problems in that area especially debate what units at the point he creates that she has a shade of any game actually uses them for drawing the crop 1 problem is that in some cases that you live in shade us the cold depends on some excellent states I they're there but shader and the fragment shader have to be linked to get us so we have to wait for the final draw tool to actually links them and this makes it just just creating him and then use is pretty comfortable but is obviously causing troubles the on we can improve that for the average case for example we can use such a linear be separate objects to avoid the linking program I don't know if that it it may just move the problem into the compiler and into the dry right and we have a test that that but I'll put it does not hence the uh fall for 99 % of the shade you can make reasonable assumption that she creation time what's runtime dependent values that we use can provide 1 of them and we can and you can create 1 Licentiate up and figure out all the game it's this corner case and just create another and in the isolates their stuttering through the tests those games that are broken upwards day problem is and this also knowing the spread of the mass effect to that very often the games do not give us the shade of what actually used so even even if it creates the open-shell shade I the time we received directed the shade up it doesn't help because it's just at the same performance-critical spot I'm so what media has done and I think what's the include developers recent work on restoring the shaders on the disk and when it gets same application is started again in our LoadOrder shows that as the what industry that catch and then a gaming creates a chill associated just match it with the hatch so there already created shade on this fixed Mass Effect 2 on the media driver but personally I think this is a fairly fairly obvious solution because the borders garbage and if this rendition of code changes so I causes problems and maybe maybe there are some nice there's 1 thing 1 was of that year that was mentioned is screwed 1st creating an out and optimize trade a for the 1st floor combining an optimized she'd in there in a separate and once the separate studies done combined optimized to adjust for basic on does anyone of you happen in a alternative ideas through storing those things and that this adds know what in that it is difficult I we can use the old assembler style a B program traders they do not suffer that much from the combined problem but it's a solution we don't really like I don't believe gives us new features on immediate parts so you prefer the use tuna salad but there are this is like the the user in the the user-visible symptom of that is that the game reasons for half a 2nd when you enemy pops around the corner field then it kills in this have 2nd anything all you know you that because you had the compiler shade of going to combine all their I this and at the end as a question was if we have like old game companies interested in that I do you mean in the sense that are all that kind of you on are you mean that the game company for example the pizzas to shaded roles 1 time the rate
later so I guess yes we have many big European companies tests the games in mind on most of the time my job we know that this has Stanford system in mind the stated so publicly because like once it once or twice a year is someone who gets banned for cheating in World of Warcraft contains all I have been using line have been banned for 4 because of that reason you even to free software and this that says no you value values in mind we detected that but the venue for that you could be paying you because you use that key what port of all tight offline the so the test so they test and the achieved system in line I believe that wealth tests they games and and achieve software online and sometimes game developers contact us and say hey we know this doesn't work in mind you know why can you fix this are not not auditing companies students but why did you do and sometimes only contact them and say hey you're doing something wrong the other fix that unfortunately never fixed my favourite problem which is infinite and not the number in the eye and traders so we added so the next point I like that the start of the does have compare instructions they're essentially compare 1 3 input values to compare the first 1 to 0 and then as such as it is essentially a condition assignment and what we wanted to intruders and is generated for each component that happens generates this up trying every the if statement and I am told that x is the that's this generates pretty ugly cold in research especially when such as statement happens to be inside into and our and media optimize the heck out of those statements and by that role in France in the mode of the probe not number semantics so venue was was 0 is not a number and then you get source 1 has through video cards because state inverts the condition because it happens to generative to remove 1 or 2 instructions and so the cop gets the nod the number right hard evidence to condition is inverted you get the wrong result in this case and technique each year this cell does not specify how low the number should be working soul from a legal point of view they are free to do that but it causes problems for us in our test cases and in some games so maybe we can you should think about the adding and specifically Julius instruction that matches the semantics of and he's conveyor maybe that that would stand with remove then the necessity for students and compilers tool to optimize those conditional assignments and would make it easier to to support floating-point specialists do gender thoughts on that like colonies making a neutral extension is always a lot of people and then have to convince the harder windows to implemented and so on but this is this is setting something to think about and the other thing I like to threading in the open shell driver and media has adoption it's and it's implemented in video travel it's turned off by the forward it's implemented in solid drive us in always stand can find it should it it's not needed authorities for the very for the the supports of the because you're working on their own way of doing that and you want to do it ourselves because for some not directed these pretty strict on at all on non-core ordering how much was clicked in open sheerly is solving needs all on synchronization for correctness anyway and here we have phone ring buffer the right to all of the lot all we have to use cheer flashed should finish or maybe a single object so we prefer the demotivating ourselves what native open-chain games with proper the lot because like on on the NVIDIA the native and that's what the main metric is and spreading which Durkheim influence on by setting an environment there were before loading the game or something like that and know what the interface looks like this and I think my defining something that has been talked about the limits mailing our mailing list completely differential concerto is out what to do with resolution changes anything precious our windows as an extra posts and ministers of that black what changed its basic things that are it allows the games to headed windows to restore the resolution when they exit the so some games just said this and then they are downplaying adjust script by calling according to CX and function of the collective process and then next summer process restores the screen resolution technically we can handle it line ourselves because we have also have some some projects and manage to screen but this is a game that that is a problem that also affects an administrative games so if you start our text Research Institute is the solution can it really flawed your screenwriter change resolution so that maybe be something to think about 4 for next when the next so road and resized by extension and finally when you book on on some dry something and you see that mind us some stupid college so these dialogs the do not so we don't have your perspective the sometimes to move notice this view we try to avoid redundant state changes and things like that the someone do not notice them sold to private developers may have a different perspective on on something so when when you notice that you doing stupid things are the about reports send me a private you may in
send it to the mailing list hours so we we can and try to fix it I'm not yet at the 1st thing we're going to change in this regard I so use of legacy context so that's that's an old problem and I know it causes performance difficulties the but rather than just complain about the state of performance 1 thing to keep in mind is that lots and other things of working Bell so what do we signed line you and even result in 9 states reckoned about much armada putting support this that the order games that have a high repeatability they work just fine so someone who is the notes on this whole machine and it's game up you can break this stock of to guarantee the me he does not purport windows for that and this year's were running running the repayable games that's doing that have the highest fences graphics it is I think a lot more important than the than running that the newest fancy game at record frame rates and people who want direct of frame rates and that's a assistance greed unity debut probabilistic evil the windows order NVIDIA binary all but if I happen to have a relationship you in my laptop and I just want to play Starcraft 11 to play warcraft or whatever that was a major vampires I can do that with the cold it's it's something physical the chips and this today and that is an achievement that should not be underestimated and it also a not a big achievement is that the money come on the most sitting in my experience works pretty well and that's he would take it for granted but that was a lot of you can imagine z is to summarize my main point is that both flying and me so like manpower especially than on parts of the cell so I believe we should focus on 1 cold and making this 1 common cold up right but rather than have a half a season 9 Implementation here have been given an implementation there have they eventually implementation here and if the if you make the common cold have worked well then they could extensive work well and the 9 games with the violence direct programs with lots of and so that's that's at least cell opinion of divine project and that's the main reason but why we problem most likely will not but I where currently do not accept that the lines that line part of their it altered 39 state TREC QA upstream because we think it's you don't it's neither sufficient nor necessary to sort the actual problem some some tests so warrant years you've seen my performance graphs I could use some help to devise a regressions especially error that you all overhead has the the net somewhat tiny benchmarks for that face the sometimes show up of of regressions and I usually radiates a few weeks before a bicycle because sometimes fixed on their own I could use some help with that but in keep in mind this is this this set of necessary for that is in all some books so sometimes I people company yeah I want to do that and then the figure out that therefore spent a day you to just to set up the tests are they they give up on that again so if you are interested you have some computers to test the song and are interested in helping out our no see knowledge is needed but you have to pay reduce compiler project obvious you send me an e-mail and also when you play games on Linux use use to get I use to call for make it repositories try to catch regression so it kind the report and early are it's much easier to fix about 5 3 days after it went into the tree rather than fix it up to the next release so that's mostly from my part I if there are any questions something now is a good time yes the and but so are so the question was if they're measurements and anywhere it likes 11 0 valent identified that is I did not test and so I don't know the but he was a i'm a mediator at the is but I think have I don't really expect to because this foreign spend what relents thus vendor are defined in this sense but the final % on except for gene years I usually not the blocking the the limiting factor so I did not test so I may be wrong but I don't think they the improve those things that are measurably anything else bias the the the the the are the question was if a Triton so Mike sufficient studio online and not no unfortunately I have not to my to my knowledge Shiites but then you aversions like anything has vicious 6 does not work properly because of that but I may be wrong and that so you can answer that in the part of and of the you and so harmonization of so but you have to find the right thing at the the all and you and what all would you know what the the so that the sum up the answer for a video and the answer
was that you can make met up after to like the fish is it because 10 working but so I need some tricks and those things that's recorded in application database but there are other issues and box and you need look around the what questions tha OK so thank you for listening and and
Schnittstelle
Explosion <Stochastik>
Einfügungsdämpfung
Nabel <Mathematik>
Mathematisierung
Automatische Handlungsplanung
Zahlenbereich
Systemzusammenbruch
Kartesische Koordinaten
Aggregatzustand
Textur-Mapping
Bildschirmfenster
Systemzusammenbruch
Term
Textur-Mapping
Puffer <Netzplantechnik>
Datenmanagement
Font
Arithmetische Folge
Spieltheorie
Bildschirmfenster
Volumenvisualisierung
Ideal <Mathematik>
E-Mail
Gerade
Feuchteleitung
Touchscreen
Schnittstelle
Bildauflösung
Softwaretest
Lineares Funktional
Interpretierer
Extremwert
Zirkel <Instrument>
Stichprobe
Bildauflösung
Spieltheorie
Dateiformat
Fokalpunkt
Rechenschieber
System F
Diskrete-Elemente-Methode
Thread
Rechter Winkel
Mereologie
Projektive Ebene
Streaming <Kommunikationstechnik>
Verkehrsinformation
Aggregatzustand
Resultante
Bit
Freeware
Aggregatzustand
Textur-Mapping
Synchronisierung
Videokonferenz
Eins
Richtung
Intel
Datenmanagement
Lineare Regression
Bildschirmfenster
Speicherabzug
Kontrollstruktur
Druckertreiber
Gerade
Benchmark
Softwaretest
Lineares Funktional
Parametersystem
Shape <Informatik>
Schlüsselverwaltung
Stichprobe
Programmierumgebung
p-Block
Kontextbezogenes System
Bitrate
Dateiformat
Dreieck
Diskrete-Elemente-Methode
Rechter Winkel
Festspeicher
Dateiformat
Programmierumgebung
Mathematisierung
Term
Physikalische Theorie
Puffer <Netzplantechnik>
Weg <Topologie>
Datensatz
Unterring
Spieltheorie
Thread
Maßerweiterung
Softwareentwickler
Hilfesystem
Schreib-Lese-Kopf
Touchscreen
Soundverarbeitung
Graph
sinc-Funktion
Chipkarte
Objekt <Kategorie>
Druckertreiber
Offene Menge
Mereologie
Wort <Informatik>
Speicherabzug
Eigentliche Abbildung
Emulator
Stapelverarbeitung
Streaming <Kommunikationstechnik>
Verkehrsinformation
Resultante
Bit
Abstimmung <Frequenz>
Prozess <Physik>
Punkt
Nabel <Mathematik>
Desintegration <Mathematik>
Versionsverwaltung
Kartesische Koordinaten
Aggregatzustand
Baumechanik
Computerunterstütztes Verfahren
BAYES
Richtung
Internetworking
Übergang
Intel
Streaming <Kommunikationstechnik>
Client
Softwaretest
PCMCIA
Code
Bildschirmfenster
Speicherabzug
Kontrollstruktur
Elektronischer Programmführer
Gerade
Bildauflösung
Schnittstelle
Benchmark
Softwaretest
Parametersystem
Shape <Informatik>
Vervollständigung <Mathematik>
Sichtenkonzept
Computersicherheit
Programmierumgebung
Quellcode
Frequenz
Teilbarkeit
Ereignishorizont
Konstante
Arithmetisches Mittel
Generator <Informatik>
Dienst <Informatik>
Diskrete-Elemente-Methode
Twitter <Softwareplattform>
Rechter Winkel
Gerade Zahl
Ganze Zahl
Projektive Ebene
Information
Overhead <Kommunikationstechnik>
Ordnung <Mathematik>
Aggregatzustand
Subtraktion
Rahmenproblem
Mathematisierung
Klasse <Mathematik>
n-Tupel
Zellularer Automat
Zahlenbereich
Graphikkarte
Zentraleinheit
Mathematische Logik
Graphikprozessor
Informationsmodellierung
Datensatz
Multiplikation
Spieltheorie
Netzbetriebssystem
Notebook-Computer
Programmbibliothek
Inverser Limes
Softwareentwickler
Bildgebendes Verfahren
Touchscreen
Soundverarbeitung
Green-Funktion
Schlussregel
Physikalisches System
Paarvergleich
Quick-Sort
Office-Paket
Chipkarte
Druckertreiber
Flächeninhalt
Debugging
Mereologie
Hypermedia
Codierung
Speicherabzug
Term
Verkehrsinformation
Resultante
Einfügungsdämpfung
Bit
Umsetzung <Informatik>
Punkt
Gewichtete Summe
Nabel <Mathematik>
Web log
Compiler
Kartesische Koordinaten
Aggregatzustand
Gradient
Client
Einheit <Mathematik>
PCMCIA
Code
Bildschirmfenster
Korrelationsfunktion
Gerade
Expertensystem
Benchmark
Schnittstelle
Serviceorientierte Architektur
Funktion <Mathematik>
Caching
Softwaretest
Addition
Vervollständigung <Mathematik>
Assembler
Kategorie <Mathematik>
Abstraktionsebene
Shader <Informatik>
Güte der Anpassung
Spieltheorie
Ruhmasse
Benchmark
Quellcode
p-Block
Bitrate
Arithmetisches Mittel
Datenfeld
Rechter Winkel
Würfel
Geschlecht <Mathematik>
Grundsätze ordnungsmäßiger Datenverarbeitung
Projektive Ebene
Ordnung <Mathematik>
Overhead <Kommunikationstechnik>
Aggregatzustand
Fitnessfunktion
Rahmenproblem
Mathematisierung
Entscheidungsmodell
Zahlenbereich
Zellularer Automat
Term
Code
Unterring
Spieltheorie
Mini-Disc
Zeitkomplexität
Luenberger-Beobachter
Installation <Informatik>
Softwareentwickler
Optimierung
Hilfesystem
Touchscreen
Beobachtungsstudie
Soundverarbeitung
Wald <Graphentheorie>
Matching <Graphentheorie>
Zwei
Physikalisches System
Binder <Informatik>
Chipkarte
Objekt <Kategorie>
Druckertreiber
Flächeninhalt
Offene Menge
Hypermedia
Mereologie
Overhead <Kommunikationstechnik>
Kantenfärbung
Mini-Disc
Compiler
Harmonische Analyse
Gewichtete Summe
Nabel <Mathematik>
Compiler
t-Test
Ungerichteter Graph
Aggregatzustand
Formale Semantik
Richtung
Netzwerktopologie
Lineare Regression
Code
Statistische Analyse
Skript <Programm>
Notepad-Computer
E-Mail
Gerade
Phasenumwandlung
Schnittstelle
Benchmark
Expertensystem
Softwaretest
Caching
Befehl <Informatik>
Sichtenkonzept
Ausnahmebehandlung
Kontextbezogenes System
Rechter Winkel
Konditionszahl
Ordnung <Mathematik>
Programmierumgebung
Fehlermeldung
Subtraktion
Relationentheorie
Mathematisierung
Open Source
Virtuelle Maschine
Lesezeichen <Internet>
Perspektive
Spieltheorie
Maßerweiterung
Offene Menge
Compiler
Resultante
Punkt
Prozess <Physik>
Computerunterstütztes Verfahren
Systemzusammenbruch
Synchronisierung
Videokonferenz
Eins
Prozess <Informatik>
Bildschirmfenster
Druckertreiber
Figurierte Zahl
Einflussgröße
Bildauflösung
Lineares Funktional
ATM
Dokumentenserver
Prozess <Informatik>
Spieltheorie
Quellcode
Ein-Ausgabe
Teilbarkeit
Diskrete-Elemente-Methode
Projektive Ebene
Overhead <Kommunikationstechnik>
Aggregatzustand
Zahlenbereich
Zellularer Automat
Implementierung
Graphikkarte
Puffer <Netzplantechnik>
Datensatz
Task
Unterring
Software
Notebook-Computer
Konstante
Inverser Limes
Zusammenhängender Graph
Eins
Softwareentwickler
Optimierung
Hilfesystem
Touchscreen
Autorisierung
Lineare Regression
Linienelement
Bildauflösung
Mailing-Liste
Physikalisches System
Ordnungsreduktion
Objekt <Kategorie>
Modallogik
System F
Druckertreiber
Touchscreen
Hypermedia
Mereologie
Mini-Disc
Verkehrsinformation
Diskrete-Elemente-Methode
Quader
Datenhaltung
Kartesische Koordinaten
Aggregatzustand

Metadaten

Formale Metadaten

Titel Wine Development Updates, Performance and the D3D9 State Tracker
Alternativer Titel Graphics - D3D9 Wine
Serientitel FOSDEM 2015
Autor Dösinger, Stefan
Lizenz CC-Namensnennung 2.0 Belgien:
Sie dürfen das Werk bzw. den Inhalt zu jedem legalen Zweck nutzen, verändern und in unveränderter oder veränderter Form vervielfältigen, verbreiten und öffentlich zugänglich machen, sofern Sie den Namen des Autors/Rechteinhabers in der von ihm festgelegten Weise nennen.
DOI 10.5446/34412
Herausgeber FOSDEM VZW
Erscheinungsjahr 2016
Sprache Englisch
Produktionsjahr 2015

Inhaltliche Metadaten

Fachgebiet Informatik

Ähnliche Filme

Loading...
Feedback