[libre-riscv-dev] TLB Update

[Luke Kenneth Casson Leighton]

---
crowd-funded eco-conscious hardware: https://www.crowdsupply.com/eoma68

On Sun, Feb 10, 2019 at 8:54 AM Luke Kenneth Casson Leighton
<lkcl at lkcl.net> wrote:

> > I have been reading that 128 entries are of a decent size if we want to
> > upgrade. I read that in a paper found here: Paper
> > <https://www.google.com/url?sa=t&rct=j&q=&esrc=s&source=web&cd=10&ved=2ahUKEwjuzPay1LDgAhWTPH0KHUBIBCQQFjAJegQICRAC&url=https%3A%2F%2Fwww.cs.rutgers.edu%2Fresearch%2Ftechnical_reports%2Freport.php%3Freport_id%3D703%26series_id%3D1&usg=AOvVaw2My8IXotmCGVV9b4D01_Id>
>
>  will take a look.

 ok, interesting, section 5.2 they have a whopping *48* hardware
threads, and, in addition (page 5) they say "GPUs do not need to run
OS code".

also, note that they say that the data throughput is usually much
higher in GPU workloads than in CPU ones, and with a lot less re-use.

 one thing that occurred to me for a software TLB that a hardware TLB
cannot really do [easily]: in the exception that handles the TLB, we
may consider loading *multiple* entries into the TLB, based on dynamic
analysis of the workload as well as past execution patterns.

this is a good find, this paper.

l.