1 Link to lecture videos

is now in the top menu bar.

2 How I show files from parallel.ecse to the class

Especially graphics stuff like PDFs. The problem is that graphics over ssh is slow.

/parclass is a git repository. There's a copy on my local laptop. I display from that.

3 OpenACC concluded

I recommend OpenACC for Programmers: Concepts and Strategies By Sunita Chandrasekaran, Guido Juckeland. Informit, Amazon. Github.

Chapter 4 is online for free as PDF and html.

We'll try a program. Note how using

export PGI_ACC_TIME=1

causes timing info to be printed.

4 Nvidia GPU and accelerated computing

This is from https://developer.nvidia.com/teaching-kits-downloads

My local copy of what I'm using is in /parclass/2021/files/nvidia/GPU-Teaching-Kit-Accelerated-Computing.zip

We'll spend several classes on this topic, speed reading through the slides.

Even though these slides are dated 2020, they have some obsolescent ideas, like explicitly copying data between host and device.

Also, some terminology is not standardized, and may be inconsistent. Even in Nvidia docs.

Today we did up through Module_2_Introduction_to_CUDA_C/Slides/Lecture-2-3-cuda-parallelism-threads.pdf .

5 Archivemount

This is today's programmer productivity tip, unrelated to parallel computing but generally useful.

Rather than explicitly extracting large zip archives or tarballs in order to read files in them, I use archivemount to create a virtual file system. This saves disk space. It doesn't stress git as much (fewer files). When reading, the I/O time is insignificantly increased. For some formats, you can even write. You can have more confidence that the zip file wasn't changed, than in a directory with perhaps hundreds of files.