The difference is what type of improvement you can make.
Intuitionistic logic enabled me invent three queues: one scales concurrency with the size of the queue; one unbounded blocking queue with mostly constant memory footprint; one priority queue that stops having logarithmic complexity under contention. I can guarantee you no amount of assembly language experience can help make such a qualitative improvement.
HPC expert will know some cool tricks. But then the cache layout changes, and you've got to learn new tricks. That's exciting and all. But to invent Barrett reduction you need something else.
You can bang on about the equality of cool tricks and the stuff like Barrett reduction. But they are not; they are qualitatively different.
Knowing how caches work guided me with assigning NUMA affinity. Knowing assembly enabled me implement the said Barrett reduction. But to devise such an application of Barrett reduction I had to know something else.
![[personal profile]](https://www.dreamwidth.org/img/silk/identity/user.png)
timeasmymeasure
no subject
Date: 2018-08-05 08:12 am (UTC)Intuitionistic logic enabled me invent three queues: one scales concurrency with the size of the queue; one unbounded blocking queue with mostly constant memory footprint; one priority queue that stops having logarithmic complexity under contention. I can guarantee you no amount of assembly language experience can help make such a qualitative improvement.
HPC expert will know some cool tricks. But then the cache layout changes, and you've got to learn new tricks. That's exciting and all. But to invent Barrett reduction you need something else.
You can bang on about the equality of cool tricks and the stuff like Barrett reduction. But they are not; they are qualitatively different.
Knowing how caches work guided me with assigning NUMA affinity. Knowing assembly enabled me implement the said Barrett reduction. But to devise such an application of Barrett reduction I had to know something else.