mozilla :: #gfx

11 Aug 2017
00:28mchangmstange: ping?
02:41mstangedvander: do we have telemetry for device pixel ratio?
02:42mstangedvander: alternatively, do you know how I can get show me telemetry of screen information that you have on ?
03:37Jerry_IRCCloudgw: ping.
03:38Jerry_IRCCloudgw: <= after rebase, should I get the r+ again?
03:50gwJerry_IRCCloud: it would perhaps be cleaner to have two separate variables, max_device_size that gets passed to the texture cache, and then max_texture_size - that way you don&#39;t need a mut at all. but yes, r+ once rebased
05:44Jerry_IRCCloudgw: r? again. There are two separate variables now.
05:54gwJerry_IRCCloud: r+ed
05:54Jerry_IRCCloudgw: thanks.
13:18katsgw: i do not (have admin rights on dev-tech-gfx). according to the mailman page it&#39;s run by milan/jrmuizel/Bas
14:39milangw: what can I do for you
14:43katswhat&#39;s this about println! causing crashes... never seen that before
15:22Gankrojrmuizel: rhunt: so the biggest issue is that modern ABIs (e.g. ARM64) distinguish between heterogeneous and homogeneous composites
15:23kvarkaosmond: current debug config
15:23GankroSo whether Opaque(size: 4, align 2) is (u8, u8, u8, u8) or (u16, u8, u8) matters
15:24GankroAnd also there are issues about how the type is treated based on argument position
15:24aosmondkvark: my build is a little different....
15:24aosmondI do disable rust debug, and use -Og
15:25Gankroe.g. Rust might choose a different &quot;apparent&quot; repr to use registers better or something
15:26GankroAlso homo-float is different from homo-int, I think (still reading the ARM spec)
15:27kvarkaosmond: thanks! trying with disabled rust
15:27Gankrolongterm (months?) eddyb thinks he can have this info exposed in a reasonable way, but currently it&#39;s a bit of a spaghetti mess down in trans
15:30GankroThere&#39;s also vague concern of, even if we had all this info, C++ making it difficult/impossible to reliably mirror what Rust says
15:34jrmuizelGankro: ah
15:34rhuntGankro: this applies to extern &quot;C&quot; as well as rust calls?
15:34Gankroyes, it&#39;s actually only in extern &quot;C&quot; since rustcall is relatively basic/lazy
15:35jrmuizelGankro: it would be interesting to put together an etherpad to keep track of issues like this so we can try to find solutions/document it for others
15:36Gankro(c ABIs are super engineered because they can&#39;t be changed, independent tools need to agree)
15:36Gankrojrmuizel: sure I&#39;ll look into doing that
15:42rhuntGankro: so if we had a simple rust struct (no enums or anything fancy) and we knew this calling convention info, would a single C struct work, or would we need one for in memory and another for crossing FFI
15:43Gankrorhunt: I&#39;m not currently clear on this, we went past 1am and wanted to sleep, feel free to ask eddyb in internals
15:48rhuntGankro: yeah I skimmed that conversation, apparently having nice cross language bindings isn&#39;t seen as that important
15:48rhuntGankro: do we have any guarantees about Option<&T> at the least?
15:49Gankrorhunt: yeah too many people use that for it to break
15:54katsmstange: ping
15:56katsmstange: unping, i think i found what i was looking for
18:52PorkepixHi, little question. Is it possible, on OS X, to gather informations to know what is triggering a switch to the dedicated discrete graphic card? Until now I sometimes did see that on websites like Google Maps. It was a bit annoying, but well, why not if it needed hardware acceleration. But with today&#39;s nightly that did broke uBlock origin, I just discovered that some ads were actually switching
18:52Porkepixthe graphic card, which is a lot more of a problem imho
18:53Porkepix(I might be on the wrong channel and would be glad to know the right one if that&#39;s the case)
18:55jgilbert_Porkepix, it&#39;s switching when webgl is activating
18:56Porkepixjgilbert_: WebGL is a pretty recent tech, right? I mean, I already had that behavior on Google Maps like 5 years ago
18:56jgilbert_possibly flash too, if you have that installed
18:56PorkepixFlash isn&#39;t installed
18:57jgilbert_webgl is about 8 years old, and Maps has been using it for quite some time
18:57PorkepixMmmh, okay
18:57PorkepixWell, actually it&#39;s pretty legit for Maps imho. It&#39;s a lot more a problem if ads just drain your power doing this kind of things
18:57jgilbert_we&#39;re working on an api to let a page ask for a low-power webgl context, but it&#39;s not implemented yet
18:58jgilbert_yeah, high-power ads are an emerging problem
18:59PorkepixIf this is an API, how websites will be supposed to use it? It&#39;s a first implementation to make it a W3C standard for every browser?
19:00jgilbert_yeah, there&#39;s standards work ongoing, and prototyping is happening in some browsers now
19:00PorkepixMmmh, this is actually a video ad
19:01PorkepixBut when you go on video websites like youtube or others, it doesn&#39;t trigger the graphic card
19:01PorkepixWhy is this one doing it :/
19:01jgilbert_different codec, probably
19:01jgilbert_I have to commute, but I&#39;ll be back in around an hour
19:01PorkepixIt looks like whatever bad practice might exist, ad providers jump on it
19:02PorkepixOkay, anyway I think I&#39;ve got most of my answers, thanks for it
19:02PorkepixAnd the main answer I guess is: there is nothging we can do?
19:42jrmuizelbotond: ping
20:38jrmuizelmystor: ping
20:39mystorjrmuizel: pong
20:39jrmuizelmystor: are you using megatron?
20:39mystorjrmuizel: I am logged in yes
20:39jrmuizelmystor: are you using it?
20:39mystorjrmuizel: Not actively
20:40jrmuizelmystor: if you log out I can start up the icecream vm
20:40mystorjrmuizel: Ive disconnected
22:21botond|homejrmuizel: pong
22:27jrmuizelbotond: already dealt with the problem
22:27jrmuizelbotond: it was a compiler bug
12 Aug 2017
No messages
Last message: 10 days and 7 hours ago