TE
科技回声
首页24小时热榜最新最佳问答展示工作
GitHubTwitter
首页

科技回声

基于 Next.js 构建的科技新闻平台,提供全球科技新闻和讨论内容。

GitHubTwitter

首页

首页最新最佳问答展示工作

资源链接

HackerNews API原版 HackerNewsNext.js

© 2025 科技回声. 版权所有。

An Apple ISA is coming

278 点作者 samps超过 9 年前

28 条评论

legulere超过 9 年前
This has been stated before: LLVM Bitcode is architecture specific. It&#x27;s dependent on the architectures ABI and can contain inline assembly. There are LLVM targets with corresponding ABIs that are nonspecific to architectures like PNaCL but Apple is not using them<p>Things Apple can do with bitcode: Produce binaries optimized for different ARM processors. Reoptimize with new versions of LLVM as soon as they are available. Use a big database of LLVM IR to tweak optimizations. Keep some optimizations secret and not even publish binaries doing them.<p>The biggest argument IMO that speaks against an Apple ISA is that they would have to rewrite tons of hand tuned assembly code.
评论 #10192025 未加载
评论 #10190843 未加载
评论 #10190781 未加载
评论 #10192666 未加载
spiralpolitik超过 9 年前
To understand Apple today you have to look to the past. Apple ended up being stuck with Motorola&#x27;s inablity to deliver faster PowerPC chips. Whole product lines were delayed, or not possible. Effectively they gave control on when they could ship new product to a third party.<p>10 years later they are now in the same position with Intel. If Intel delays the next version of its product line by six months then Apple has to put things on hold. This is bad for a company like Apple as it could cause them to miss out on potentially lucrative periods (back to School, the Holiday season etc).<p>Ultimately I suspect in the very near term we will see Apple move off Intel, first for the laptops. LLVM IR would fit this strategy better than fat binaries as Apple would not have to wait until developers recompile. They can have the entire App Store available on day 1 of a product release.
评论 #10191616 未加载
评论 #10191247 未加载
评论 #10191347 未加载
评论 #10192046 未加载
评论 #10192104 未加载
评论 #10192067 未加载
评论 #10191491 未加载
评论 #10191255 未加载
评论 #10194867 未加载
gchadwick超过 9 年前
It is a common theme in computer architecture for someone to say, &#x27;look I have an awesome new architecture all you need is a clever compiler to make it work&#x27;. None of them really pan out (other than in specific applications, you could argue GPGPU is as an example, though that appeared more by accident than by design).<p>Apple would need a very good reason to produce their own ISA. Sure they like to do many things in house but they don&#x27;t do everything themselves. The resources required to produce and support a whole new ISA are a major investment, they&#x27;re only going to do it if in the long run is cheaper than paying for an ARM architecture license. I just don&#x27;t see a solid argument that a custom ISA and shiny new compiler would give them much (if indeed anything).<p>Whilst they may be shipping LLVM IR for the watch apps rather than ARM code I think this is just so they can target the compile for a specific processor. Each one has its own performance quirks and especially in such a power sensitive environment it would make sense to do specific targetting.
评论 #10190411 未加载
评论 #10191573 未加载
评论 #10190529 未加载
评论 #10192157 未加载
评论 #10190489 未加载
geoff-codes超过 9 年前
I don&#x27;t think this holds water at all.<p>The crux of the article is:<p>&gt; ...one last gap remains in the middle of this stack of system exclusivity: Apple licenses the instruction set architecture for its mobile devices from ARM.<p>But Apple already not only designs the own SoCs independently already, they regularly add their own opcodes to the ARM instruction sets they license, as they see fit.<p>The alternative to <i>not</i> licensing from ARM, even if they &quot;invented their own ISA&quot;, would be to pay an exorbitant sum in royalties to ARM <i>and</i> every other patent-holder whose technology they might dare use in their chip. So paying ARM for their technology in one go just makes the most economical&#x2F;legal sense.
haberman超过 9 年前
It&#x27;s widely disputed that LLVM bitcode is actually ISA-agnostic. There is a HN comment quoting Chris Lattner as saying that CPU independence isn&#x27;t really the point of bitcode apps (I&#x27;d find it now but I&#x27;m on mobile). The thought is that it has a lot more to do with the ability to re-optimize.
评论 #10190576 未加载
raverbashing超过 9 年前
Though the arguments are interesting, I&#x27;m not convinced<p>Sure, the semantic gap exists. But ARM and x86 have evolved and have overcome a lot of difficulties.<p>People like to bash x86 but it has a big advantage: it&#x27;s compact. ARM Thumb is compact but not so much.<p>Also remember how big the last &#x27;new ISA&#x27; (Itanium) success was?<p>Compiler and processor front-end beat &quot;perfect ISA&quot; today
评论 #10190396 未加载
评论 #10190507 未加载
评论 #10190498 未加载
评论 #10194251 未加载
zvrba超过 9 年前
&gt; a traditional von Neumann ISA like ARM incurs a semantic gap; the architecture wastes time and energy rediscovering facts that the compiler already knew.<p>So thought the designers of Itanium too; it turned out that the compiler doesn&#x27;t know <i>sufficiently</i> much for an architecture like Itanium.<p>Were there any significant advances in compiler technology since then that would make it worthwhile to experiment with a new ISA?
hexscrews超过 9 年前
This seems a bit like click bait. It holds no substantial information. They might as well have summed up the article with, One day, apple will make their own processor. That&#x27;s all it says.
评论 #10190500 未加载
评论 #10190555 未加载
评论 #10190844 未加载
protomyth超过 9 年前
I don&#x27;t think they&#x27;ll produce their own ISA. They have an architectural license from ARM, so why bother.<p>But, Intel doesn&#x27;t exactly produce chips that are helpful to Apple. Since Apple switch, Intel has gotten rid of third-party chipsets. This removed a lot of customization options and basically made life easier for Intel since they produce a fixed set of chips and you have to take them. Also, Intel&#x27;s market differentiation of chip features probably doesn&#x27;t help.<p>Apple wants to provide a custom experience, and Apple building their own PC-class ARM chips will allow that.<p>[edit: also Intel&#x27;s paying people to produce Macbook Air clones probably didn&#x27;t help]
评论 #10191910 未加载
ChuckMcM超过 9 年前
I&#x27;m read this wondering if the author had any remembrance or understanding of the original CHRP&#x2F;PReP fiasco (<a href="https:&#x2F;&#x2F;en.wikipedia.org&#x2F;wiki&#x2F;AIM_alliance" rel="nofollow">https:&#x2F;&#x2F;en.wikipedia.org&#x2F;wiki&#x2F;AIM_alliance</a>). That was a painful time for Apple. Given the nature of how ARM operates, and the fact that Apple has has a full ARM license (so could, at their leisure add special sauce in the instruction set if needed) that they would do a new ISA, they already did, x86-64 =&gt; ARM. The question for me is whether or not their desktop&#x2F;laptop series moves that way or not. I&#x27;ve said for years a 12&quot; MacBook Air running IOS would put a huge dent in the Chromebook market, and a 12&quot; ARM based Air running IOS? Well that would be a pretty obvious move to me.
rkangel超过 9 年前
This article points to another bit of the technology stack that Apple doesn&#x27;t own - LLVM IR.<p>LLVM is open source and therefore doesn&#x27;t require licensing unlike the ARM Instruction Set, but it&#x27;s another thing they don&#x27;t perfectly control and they&#x27;re happy with that.<p>Developing a new ISA would be extremely expensive, and they&#x27;d have to have a really good reason for doing it. The post doesn&#x27;t suggest why it would be beneficial, merely extrapolates a pattern.
评论 #10191079 未加载
评论 #10190903 未加载
评论 #10196511 未加载
tbrock超过 9 年前
Apple has owned part of ARM holdings as part of a joint venture between them and Acorn for decades now. That was one of the first pieces of the widget they had. Joke is on everyone making phones and servers: you are paying Apple already.<p>From Wikipedia:<p>&quot;The company was founded in November 1990 as Advanced RISC. Machines Ltd and structured as a joint venture between Acorn Computers, Apple Computer (now Apple Inc.) and VLSI Technology.&quot;
评论 #10197760 未加载
joosters超过 9 年前
I can&#x27;t see this as a definitive sign that Apple are going to introduce a new ISA. Instead, this gives them flexibility. They can switch ISAs (e.g. to a new ARM revision, it doesn&#x27;t have to be an Apple-specific one though) without authors having to recompile their code for all platforms. It also allows Apple to support multiple ISAs without code bloat - the app store can download the correct binary to your phone, just like they now can send only the correct size images.<p>I don&#x27;t know how abstract the LLVM IR is - can you take IR and compile it to two wildly different ISAs, (say) x86 and ARM, and get full optimisation on both? Or is it more limited, e.g. allowing you to compile to ARM version x and ARM version y (e.g. if version &#x27;y&#x27; supports some new SIMD instructions).
评论 #10190542 未加载
clord超过 9 年前
I think the Mill team is a good acquisition target for Apple... Apple could swiftly switch their whole stack over and gain remarkable benefits. I can&#x27;t think of another company that would even be able to switch to the mill line.
omarforgotpwd超过 9 年前
Given the effort put into engineering support for bitcode in iOS 9, it&#x27;s clear that the processor&#x27;s instruction set is definitely going to change at some point. The only question is when. I wouldn&#x27;t be surprised if these new processors were unveiled about 8 and a half hours from now.
评论 #10190474 未加载
hajile超过 9 年前
Apple doesn&#x27;t want their own ISA. MIPS was going for a song just a couple years ago and Apple decided not to buy. That was the best shot they&#x27;ll have at their own ISA (they could go with RISCV, but they wouldn&#x27;t own it).It is my belief that they got what they wanted by forcing ARM to switch to the ARMv8 ISA.<p>New microarchitectures take 4 years or so to design. ARM announced the new ISA in 2011 and didn&#x27;t have a shippable product until 2015 which is very typical. All the other implementers (eg. Qualcomm) have also not been able to ship until now (Qualcomm&#x27;s custom Kryo doesn&#x27;t hit until later this year). Apple shipped a better product in 2013 than A57 is today (ARM doesn&#x27;t catch up until A72 later this year). To my knowledge, a licensee had never shipped a new ISA before the ISA designer up to this point. How did they get a chip designed, validated, tapped out, produced, integrated, and shipped out in 2 years?<p>I believe that Apple looked into purchasing MIPS or designing a custom ISA, but was put off by the costs and headaches associated with moving ISAs (having already done this with the change from POWER to x86). Instead, they design an ISA that is incredibly close to MIPS and start implementing a micro-architecture. Once they reach the stage where they must make a decision about which ISA, they tell ARM to use their ISA or they will move to MIPS. This head-start also<p>ARM is already somewhat threatened by Android having first-class support for MIPS. Having such a big player switch would be extremely threatening to them. The result would be an immediate caving. ARM would need to publish the ISA, but Apple would have a couple year head-start on implementing it (this head-start also puts Apple in a good competitive position relative to Android phone manufacturers). The rest is observable history.<p>This may not accurately represent what really caused this series of events, but it does explain why Apple got a good chip out before ARM could release a bad one (ARM couldn&#x27;t even get a smaller, easier chip out the door). It also explains why all the other chip companies hint at their surprise at Apple&#x27;s early launch.
hyperpallium超过 9 年前
Getting rid of accumulated cruft would simplify decode logic. Customize it for iOS. Siliconize common functions.<p>Apple doesn&#x27;t have to worry about standards, backcompatibility or adoption. If they have noticed unneccesary inefficiencies, they can fix them.
Tloewald超过 9 年前
Very insightful little post. In theory Apple could start evolving their ISA over time alongside everything else because LLVM gives them an abstraction layer. (Everyone else could do this too of course.)
Quequau超过 9 年前
I&#x27;m pretty certain that lots of folks here are looking at this from an unproductive perspective.<p>Robert Colwell from DARPA presented a talk at HotChips 2013 which was focused on post &#x27;Moore&#x27;s Law&#x27; technologies and he brings up specialized ISAs.<p>Looking at the potential of Apple releasing chips with new ISAs from this perspective seems to me to make a lot more sense (to me at least).<p><a href="https:&#x2F;&#x2F;www.youtube.com&#x2F;watch?v=JpgV6rCn5-g" rel="nofollow">https:&#x2F;&#x2F;www.youtube.com&#x2F;watch?v=JpgV6rCn5-g</a>
nikdaheratik超过 9 年前
Thought provoking, but I disagree with pretty much every conclusion. If there is anything they <i>really</i> need they can likely get ARM or Intel to put it in there, so building an ISA from scratch isn&#x27;t going to gain them much. It&#x27;s the same reason why they aren&#x27;t going to become a cell-phone carrier even though that&#x27;s another part of the vertical stack that they could try to get into: the cost&#x2F;benefit makes it not worth it.
fulafel超过 9 年前
This &quot;get freedom from binary compatibility by shipping binaries as compiler IR&quot; concept is a venerable one and has been done many times with VLIW machines:<p>See here, <a href="https:&#x2F;&#x2F;scholar.google.com&#x2F;scholar?hl=en&amp;q=A+Technique+for+Object+Code+Compatibility+in+VLIW" rel="nofollow">https:&#x2F;&#x2F;scholar.google.com&#x2F;scholar?hl=en&amp;q=A+Technique+for+O...</a><p>(Not to mention earlier examples like AS&#x2F;400, P-code etc)<p>I think the Mill was doing this too?
abalone超过 9 年前
The key sentence is: &quot;the [current] architecture wastes time and energy rediscovering facts that the compiler already knew.&quot;<p>If Apple does something here it&#x27;s going to be for the watch, not Macs. Pushing the envelope of efficiency for the watch is where it becomes worth it to make this kind of (otherwise insane) investment. It&#x27;s also a relatively simpler stack, so more feasible.
dsjoerg超过 9 年前
In case you were wondering, like I was: <a href="http:&#x2F;&#x2F;google.com&#x2F;search?q=what+is+an+isa" rel="nofollow">http:&#x2F;&#x2F;google.com&#x2F;search?q=what+is+an+isa</a>
hitlin37超过 9 年前
trying to understand a bit more about bitcode concept. Since a developer submits a bitcode for apple watch, does that mean he can&#x27;t optimize his own app for performance?<p>FWIK, on android you can still optimize your app at assembly level and i think that&#x27;s what motivates developers at times. Remember the iphone camera high speed shot app, that was hand coded to be fast on iphone.
评论 #10193044 未加载
Aissen超过 9 年前
We&#x27;ll see an Apple GPU IP in iPhones&#x2F;iPads before a new ISA.
frozenport超过 9 年前
Why didnt Apple just ask authors to submit source code?
评论 #10195062 未加载
评论 #10192162 未加载
em3rgent0rdr超过 9 年前
LLVM &quot;owned&quot; by Apple? Can&#x27;t really &quot;own&quot; open-source.
评论 #10190587 未加载
评论 #10190439 未加载
评论 #10190634 未加载
评论 #10190422 未加载
herp_derp超过 9 年前
For Apple to develop an ISA they would have to develop a CPU. Something they don&#x27;t do inhouse.<p>Up until now the CPUs have been designed by ARM. Cores like Swift for example were not developed at Apple, but rather at ARM.<p>The likelihood of ARM releasing a chip into the wild with a non-ARM ISA is not all that great, since the ARM ISA is what ARM makes all of its living from.<p>Until the day where Apple is capable of creating a CPU by themselves there will be no Apple ISA.
评论 #10191583 未加载