2 ms·
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 -- tha
by ychen306 8y ago
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.
- StillBored 8y agoThere are a ton of domain specific languages (SQL, GLSL, etc), and I don't have a problem with those, they have a place. The problem is that we already have too many languages trying to be generic, and languages like C++, and openMP can be wrangled into nearly any programming paradigm in common use and the results tend to also be significantly faster. If your talking about performance, the minimum baseline requirement should be running faster than a language in common use, say C++. There are a few languages that might take this (Cuda/OpenCL) but they also tend to be somewhat domain specific. Similarly verilog/VHDL can produce fast results. There have been calls for languages that can express parallelism better and are both expressive, and safe for decades. But what we have actually gotten haven't been better in most objective measures. What we have gotten are a pile of "scripting like" languages with very similar characteristics (tcl, ruby, perl, python, javascript, PHP, lua, go, applescript, etc).