tag:blogger.com,1999:blog-8221886912876606828.post8123841414015028742..comments2023-06-23T00:44:05.938-07:00Comments on Elided Branches: Parallelism and the Limits of LanguagesCamille Fournierhttp://www.blogger.com/profile/05090020862261633248noreply@blogger.comBlogger6125tag:blogger.com,1999:blog-8221886912876606828.post-36989852801299131302012-05-03T09:07:58.541-07:002012-05-03T09:07:58.541-07:00This comment has been removed by a blog administrator.moroccohttps://www.blogger.com/profile/12956000163730510893noreply@blogger.comtag:blogger.com,1999:blog-8221886912876606828.post-27281910763467348912012-04-29T16:18:22.681-07:002012-04-29T16:18:22.681-07:00Since the title is Parallelism and the Limits of L...Since the title is Parallelism and the Limits of Languages, I have to ask - have you used Haskell?<br /><br />We "sacrifice" pointers in languages that provide garbage collection. In order for the language to provide better parallelism we need to sacrifice unrestrained side-effects.Craighttps://www.blogger.com/profile/15806804843602653454noreply@blogger.comtag:blogger.com,1999:blog-8221886912876606828.post-30125885272021956572012-04-24T21:43:25.901-07:002012-04-24T21:43:25.901-07:00Hey,
very nice blog!I’m an instant fan, I have boo...Hey,<br />very nice blog!I’m an instant fan, I have bookmarked you and I’ll be checking back on a regular.See u.<br /><br /><a href="http://www.apexical.com/soil-and-stain-release.htm" rel="nofollow">Stain Release</a>apexical apexicalhttps://www.blogger.com/profile/00730732343311315355noreply@blogger.comtag:blogger.com,1999:blog-8221886912876606828.post-22986685752148311842012-04-10T05:20:31.999-07:002012-04-10T05:20:31.999-07:00The idea is sound, but as far as I know there are ...The idea is sound, but as far as I know there are already quite a few very smart people at e.g. Intel and IBM working for quite a while on automatically detecting opportunities for parallelization in byte code with only some success.<br />Unfortunately this is a non-trivial problem as a lot of parallelization scope is on application level and hard to impossible to detect on byte code level, so I guess we'll have to wait for the VM magic for a little longerErich Eichingerhttps://www.blogger.com/profile/01777569034256321263noreply@blogger.comtag:blogger.com,1999:blog-8221886912876606828.post-51417983736878304642012-04-10T01:55:08.946-07:002012-04-10T01:55:08.946-07:00Bravo, you've written this idea down very clea...Bravo, you've written this idea down very clearly. I've also thought about this and I agree that this is key. A sufficiently expressive language combined with a VM or JIT could slice and dice the code so that it uses exactly the right batch size, number of threads, and even internal data representation for the best performance and occupancy of the target CPU/GPU cores. It could even be tuned at run-time based on other system load. This would be incredibly nice to have.Wladimirhttps://www.blogger.com/profile/13282912869380734925noreply@blogger.comtag:blogger.com,1999:blog-8221886912876606828.post-29866181282950634342012-04-07T16:18:37.792-07:002012-04-07T16:18:37.792-07:00The question in my mind is whether there is more b...The question in my mind is whether there is more benefit in commoditizing parallelism in the VM than in languages - recognizing that both will probably be inadequate for many problems - your last paragraph says as much. The quote from Aria Haghighi above doesn't invalidate the language approach as much as say it wasn't enough for his set of problems. Indeed, one could argue that the VM, being even further away from the problem space than the language, has less of a chance of being more successful.<br /><br />The success of the JVM's commoditization of memory management is an interesting case, but I think indicative that memory management is a "simpler" problem (but not to underestimate the achievement!) than that of the concurrency problem.<br /><br />As usual, thanks for the thought-provoking piece.Mike Mhttps://www.blogger.com/profile/11490344190768803850noreply@blogger.com