Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

David Patterson Says It’s Time for New Computer Architectures and Languages

spectrum.ieee.org

Post reply on HN
Print view View original

131–140 of 204 posts

Re: David Patterson Says It’s Time for New Computer Architectures and Languages

#131

Post by yuushi »

munificent wrote: post #79
amckinlay wrote: post #71

I want a language where you can express time and constraints on time within the language and the type system. I want to be able to guarantee program correctness, pre-calculate fine-grained energy usage, optimize for energy/power-saving mode usage, interface with asynchronous signals, and whatnot -- all with respect to program execution at the ISA-level within some hard real-time constraints. Compilers make optimizati…

Part of the reason most languages obscure this is because it's a moving target. If a language let you say, "this chunk of code here should run in 7 cycles", what happens when a new optimization finds a way to reduce that, or a new architecture comes up where that operation gets slower but lots of others get faster? I'm not arguing against your desire, just explaining that it's not unreasonable that we're where we are…

This could only ever be doable with extremely simple architectures anyway. Off the top of my head, add in just one of branch prediction, micro-op fusion, out-of-order execution, pipelines and pipeline stalls, or cache misses, and this becomes impossible. Of course, this assumes you even know which CPU you are targeting its specific instruction latencies.

That's already an extremely niche set of processors. Further, the number of bits of code you're likely to care about this kind of extremely precise timing for, you'll either examine the emitted assembly, or just hand-write the ASM yourself.

It seems like a huge amount of effort for an extremely niche scenario. Remember, the ISA is still just an abstraction, after all.

Re: David Patterson Says It’s Time for New Computer Architectures and Languages

#132

Post by henrikeh »

amckinlay wrote: post #71

I want a language where you can express time and constraints on time within the language and the type system. I want to be able to guarantee program correctness, pre-calculate fine-grained energy usage, optimize for energy/power-saving mode usage, interface with asynchronous signals, and whatnot -- all with respect to program execution at the ISA-level within some hard real-time constraints. Compilers make optimizati…

This is one of the most actively researched fields in computer architecture/engineering. Hard real-time requirements has been been extensively researched for ISAs, compilers, network-on-chips, concurrency models etc etc.

Personally, I was briefly involved in the T-CREST[0] project where a complete ISA, network on a chip and compiler where devolved in tandem to support deterministic timing and well understood worst case execution times (WCET).

As an example, the ISA explicitly defines multiple caches with different write-rules. So for example the stack cache is write-back to lower WCET.

The whole project is available as a VM image[1] where the HDL can be compiled to an emulator and/or synthesized to hardware.

[0]: http://www.t-crest.org/page/overview [1]: http://patmos.compute.dtu.dk

Re: David Patterson Says It’s Time for New Computer Architectures and Languages

#133

Post by ndesaulniers »

philipkglass wrote: post #7

Are there languages that have first-class support for representing/optimizing memory hierarchy characteristics? Optimizing C compilers, for example, may have extensions to force specific alignments: https://software.intel.com/en-us/articles/coding-for-perform... But I'm not aware of languages where e.g. declaring alignments is part of the base language. Awareness of L1, L2, L3 cache characteristics, plus NUMA nodes,…

XLA, the JIT compiler in tensorflow could hide this from you. It's a thing of beauty. It can vectorize along multiples of the memory higherachies of CPUs, GPUs, and the family of TPUs.

Re: David Patterson Says It’s Time for New Computer Architectures and Languages

#134

Post by ychen306 »

StillBored wrote: post #61

How about instead of inventing new languages we just pick a couple and tell people that if care at all about performance not to use the rest. I would start by throwing away all the interpreted/GC'ed languages because even after decades of massive effort they are frequently lucky if they even manage to keep up with unoptimized C.. But that really isn't the problem, the problem is that they cannot be debugged for perfo…

We need new languages; at the very least new ways to approach to optimization. We are at the point where even C doesn't give you the performance you want -- that's why people invented things like Halide, XLA (Tensorflow), etc.

Re: David Patterson Says It’s Time for New Computer Architectures and Languages

#135

Post by learc83 »

chubot wrote: post #24
philipkglass wrote: post #7

Are there languages that have first-class support for representing/optimizing memory hierarchy characteristics? Optimizing C compilers, for example, may have extensions to force specific alignments: https://software.intel.com/en-us/articles/coding-for-perform... But I'm not aware of languages where e.g. declaring alignments is part of the base language. Awareness of L1, L2, L3 cache characteristics, plus NUMA nodes,…

Sequoia is what I thought of at first (mentioned in sibling comment), but Scala has something related: http://scala-miniboxing.org/ldl/ Basically you can write an algorithm that is independent of the exact data layout, and then the compiler has some flexibility to work with. I haven't worked with it all, just read the paper awhile ago. But I'm pretty sure they did something with SoA - AoS transformation (structure of…

One of the differentiating features of Jonathan Blow's language, Jai, is supposed to be AoS/SoA transformation. I wasn't aware other languages were experimenting with this as well.

I thought it was something fairly specific to game programming, but I guess it shouldn't be surprising that other performance sensitive domains might also find it useful.

Re: David Patterson Says It’s Time for New Computer Architectures and Languages

#136

Post by chubot »

api wrote: post #66

Here's something that seems shockingly under-explored to me: languages that incorporate relational data structures natively. We have SQL of course, but SQL is not a general purpose language and is (intentionally) often not Turing-complete. I'm imagining something like Go, Ruby, JavaScript, or Rust with native tables, queries, and other SQL-ish relational data structures. The long term goal would be to kill the age-ol…

Data frames in R give you the relational model with a Turing complete language. I think of it as a better SQL, at least for analytics (as opposed to transaction processing).

Besides its data structures, R is surprisingly similar to JavaScript -- it's has Scheme-like semantics but C-like syntax.

For certain problems, it's a pleasure to program in. You don't have to go through "objects" to get to your tables; you just have tables!

The examples here might give you a flavor:

https://cran.r-project.org/web/packages/dplyr/vignettes/dply...

R itself has a lot of flaws, but the tidyverse [1] is collection of packages that provide a more consistent interface.

FWIW I agree that more mainstream languages should have tables natively. It is an odd omission. Somehow they got coupled to storage, but they're also useful for computation.

[1] https://www.tidyverse.org/

Re: David Patterson Says It’s Time for New Computer Architectures and Languages

#137

Post by dnautics »

stellalo wrote: post #11

Well, there are some interesting new languages out there already: https://julialang.org/ In fact Moore’s law must have been the reason why Python is what it is today: “Who cares if it’s slow, tomorrow’s computers will be faster!”

I would say erlang is an interesting new language, except that it's not new. The idea of a functional language that is just very careful about its impurities and has a clever way of managing state without spinlocks or mutexes is a great thing. Beam languages have almost no footguns and programming them feels like programming with inflated gutterball balloons. A chip architecture that was highly optimized for the beam would be great.

Re: David Patterson Says It’s Time for New Computer Architectures and Languages

#138

Post by ssvss »

graycat wrote: post #123

IMHO, stochastic optimal (SOC) control is a framework that is much more powerful than anything like current artificial intelligence, machine learning, deep learning, etc. and, in particular, a much better version of something like actual learning. The difference is like going from gnats to elephants, like a bicycle to the star ship Enterprise. A good application of SOC looks not just "intelligent" but wise, prudent,…

What book would you recommend to know more about SOC?

Re: David Patterson Says It’s Time for New Computer Architectures and Languages

#140

Post by ludwigschubert »

eslaught wrote: post #121
eslaught wrote: post #117

Earlier quoted context omitted.

(Disclaimer: I work on Legion) Sequoia has been largely superseded by its spiritual successor Legion [1], another programming system by Alex Aiken. Legion doesn't focus so much on low-level memory hierarchies, but it does a very, very good job of scaling to very large machines, and taking advantage of heterogeneous processors such as GPUs. It is also incomparably better at working with dynamic programs, whereas Sequo…

P.S. Shameless plug: If you're interested in working on this sort of stuff, we're hiring for full-time staff scientists as well as interns. Drop me a line! (Contact in profile.)

3rd party vouch for shameless plug: I was a RA in a minor role at this team and these are some of the friendliest and most approachable people I've had the pleasure to work with in academia. (Hi Elliot! \o/)
Post reply on HN