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。

A novel data-compression technique for faster computer programs

75 点作者 pps大约 6 年前

6 条评论

foota大约 6 年前
Interesting paper. The original concept is at <a href="http:&#x2F;&#x2F;people.csail.mit.edu&#x2F;sanchez&#x2F;papers&#x2F;2018.hotpads.micro.pdf" rel="nofollow">http:&#x2F;&#x2F;people.csail.mit.edu&#x2F;sanchez&#x2F;papers&#x2F;2018.hotpads.micr...</a>, and the follow up paper the article is written about is <a href="https:&#x2F;&#x2F;people.csail.mit.edu&#x2F;poantsai&#x2F;papers&#x2F;2019.zippads.asplos.pdf" rel="nofollow">https:&#x2F;&#x2F;people.csail.mit.edu&#x2F;poantsai&#x2F;papers&#x2F;2019.zippads.as...</a>.<p>Afaict the gist of the hotpads paper is this:<p>Dump the existing way we organize processor caching, instead make it look like generational gc where objects can move between different caches, and pointers can be re-written to point to the new cache level that the pointee is in.
评论 #19953818 未加载
sp332大约 6 年前
Rob Matheson writes for MIT&#x27;s Technology Review, which has a reputation for massively over-hyping anything done by MIT researchers. Just take the performance claims with a grain of salt until someone who doesn&#x27;t work for MIT writes about it.
kannanvijayan大约 6 年前
This was interesting to read. It seems like a sophisticated, hardware-enabled generational GC on top of a managed heap.<p>It reminds me somewhat of an optimization that my colleague Brian Hackett implemented on the Spidermonkey JS engine, which would use runtime-type-tracking infrastructure to discover objects that had specialized constituents (i.e. a slot was always a boolean, and had never not been a boolean).<p>The system would notice this and then transition the object to a new layout with non-value-boxed entries where appropriate. Of course this included de-optimizations hooks to allow objects to transition back to the general &quot;boxed&quot; layout if mutations to the object de-specialized the slot.<p>This technique delivered some significant performance improvements when it came to computationally heavy, type-stable code (such as what we find in the Octane benchmarks). It wasn&#x27;t as effective on type-unstable code.<p>I&#x27;d expect that for most traditional standalone programs written in a statically typed, managed language (e.g. Java) - this sort of approach has a lot of promise.
_wmd大约 6 年前
Paper: <a href="https:&#x2F;&#x2F;people.csail.mit.edu&#x2F;poantsai&#x2F;papers&#x2F;2019.zippads.asplos.pdf" rel="nofollow">https:&#x2F;&#x2F;people.csail.mit.edu&#x2F;poantsai&#x2F;papers&#x2F;2019.zippads.as...</a><p>Seems this cannot be read without first understanding &quot;object based memory hierarchies&quot;, which is new to me. This coauthor&#x27;s page has a ton of related papers: <a href="http:&#x2F;&#x2F;people.csail.mit.edu&#x2F;poantsai&#x2F;" rel="nofollow">http:&#x2F;&#x2F;people.csail.mit.edu&#x2F;poantsai&#x2F;</a>
jdnier大约 6 年前
This reminds me of &quot;RAM Doubler&quot; from the mid 90s on 68030&#x2F;40 Macs. <a href="https:&#x2F;&#x2F;tidbits.com&#x2F;1996&#x2F;10&#x2F;28&#x2F;ram-doubler-2&#x2F;" rel="nofollow">https:&#x2F;&#x2F;tidbits.com&#x2F;1996&#x2F;10&#x2F;28&#x2F;ram-doubler-2&#x2F;</a><p>I do remember the joy have doubling 16MB of RAM to 32MB.
评论 #19954220 未加载
Digit-Al大约 6 年前
What happens if you&#x27;re using a functional language?