Automatic SIMD Vectorization for Haskell [pdf] | Hacker News Reader

Automatic SIMD Vectorization for Haskell [pdf]

leafpetersen.com

86 pointsmiga18 comments

I still think it's truly awesome that I can just hapilly write my code in Haskell on a very high level, while in the meantime other people will hack on the compiler and core libraries and magically make my programs run faster!

Seriously, thank you for doing research in this area. It will put us forward.

Haskell is all about the lazy.

You can apparently blame C for the belief that a programmer should have to optimize code themselves.

Though I'm waiting for the itanium/ILP optimized haskell compiler. It's one of the few languages that tracks side effects enough that it seems possible to actually build the "sufficiently smart compiler" that Intel always claimed would materialize to make itanium a good idea.

What would you blame them for? Beating the crap out of the alternatives?

"Intel Labs Haskell Research Compiler" ... !!!

So it's going to be throttled on AMD CPUs?

Why? My impression is that it isn't used outside of intel.

http://en.wikipedia.org/wiki/Intel_C%2B%2B_Compiler#Criticis...

It's a reference to that incident where intel's C compiler was checking the vendor strings to turn on certain optimized pathways.

Relevant to the discussion about targeting LLVM vs. C in another thread (https://news.ycombinator.com/item?id=6096965), Section 2 says they compile via C, but through an intermediate language, Pillar, that adds support for second-class continuations, precise garbage collection, and a few other things. Here's an older paper on Pillar: http://www.leafpetersen.com/leaf/publications/lcpc2007/pilla...

Why do always the best link on HN on high performance, non-trivial compilers or parallelism get little to no comments?

It did get voted to the front page, so there is interest, but very few readers are able to add anything useful.

People write comments when they feel that they have useful contributions to make. It is much easier to notice that you have nothing worthwhile to say about a vectorizing compiler than about, say, the NSA.

One way to get around this is to go meta, e.g. to discuss the commenting habits of others instead of the submission.

I went meta myself, sorry about that.

No apology needed, I was trying to make fun of myself, not you.

And a link NOT pointing to scribd:

http://www.leafpetersen.com/leaf/publications/icfp2013/vecto...

The submission link points to that, the `[scribd]` is a separate mirror added automatically by HN.

Ya, its always tough to figure that out. Design fail.

They also changed the title of the post. It was: "Intel Labs Haskell Compiler: Automated..."

Reply on news.ycombinator.com