hmaarrfk

That’s a large support matrix. Congratulations.

Do you link to the shared object files? Or do you use subprocess?

What version of ffmpeg does this use? Last I tried torch tools used really outdated version of ffmpeg at the time of their release.

From the same developer ;)

Thanks for sharing. That was a great read

When you are trying to design high performance IP, you are often trying to ensure that your design is mathematically correct, inputs and outputs are matching a complicated 100 page specification. You are also trying to parse out the minimum set of workable requirements for "version 1" all with fitting into utilization constraints that ultimately are undefined.

Your mindset is really split. "Building up a software dashboard" to visualize your test results is really the last thing on your mind. You definitely don't want to be building the dashboard for all your customer's platforms.

Having somebody (a company) help on this front is really useful.

As a non-website designer, I used to think the same of tools like netlify, but they seem to be popular as ever, especially in a collaborative workspace when you need to handoff a project from one team to the next.

Thank you for tackling this critical problem for logic designiners. I think the tools available are much too old for fast paced workflows.

From my experience attempting to get a similar workflow down for my company:

I tried to use verilator a while back but ultimately I couldn't because it didn't have same constraints in the verilog language features that I was going to use in production. It doesn't even matter who was missing a feature, verilator or the proprietary tool, it was just about getting them to be same that caused the cognitive dissonance that I didn't want to deal with.

I ultimately decided to move away from verilator and use the clunky proprietary tools since it was what would be used in production. Getting "verilator compatibility" seemed like a "nice to have".

Second, the a winning local-first framework of verilator wasn't really established. You show in your example running a test from the yaml file using what looks like a bash script. Even as an experienced programmer who knows bash and sh well, I still find it very hard to write complex thoughts in it. The last high level attempt I found to bridge this gap is likely https://www.myhdl.org/ I don't know them personally, but it seemed like they had some very good thoughts on what makes writing good hardware level tests good. I think it would be worth reaching out to them if you haven't already.

The one thing that even more critical was a way to run our tests locally. The 10-20 seconds it takes to start a docker image (best case) in the cloud is really frustrating when you are "so close to finding a bug" and you "just want to see if this one line change is going to fix it". Once we got our whole pipeline going, it would take 1-6 minutes to "start a run" since it often had to rebuild previous steps that cache large parts of the design.

So I think you will want to see how you can help bring people's "local's first" workflows slowly into the cloud. Some tools (or just tutorials) that help you take a failing test, and run it locally and on the cloud will be really good especially as you get people to transition!

the other thing i notice from the story was that an update on something considered mission Critical was not given an update on within 24 hours.

IT should have volunteered the info regarding how far back in the backlog this was classified as soon as that prioritization was made. "Behind 14" and with many people on the testing side occupied is obviously not going to help with "layoff level priority".

To me, the classification of "enhancement" just doesn't seem to capture the urgency.

https://www.bugsnag.com/blog/bug-day-460m-loss/

It made me laugh! And cry inside

This is a story where a one line change of a hardcoded value actually went well.

I could image a scenerio where somebody stored the number of months of backlog as a 2 bit value. 0, 1, 2 or 3, you know, to be smart and clever. This may not appear as a problem during testing because it may be hidden many layers down, in some downstream service that is untested. Maybe in some low code automation service....

Changing it to 4 would mean the backlog is 0. Who knows what the consequences might be. Would that service go and cancel all jobs in the production queue? Would it email all customers mentioning their stuff is cancelled?

I get that this is a seem gly easy change, but if a change of policy is expressed to the software team as an urgent problem, this seems like the management team needs better planning, and not randomly try to prioritize issues.....

Thanks for this link.

They also seemed to have avoided libraries like numpy.

In my mind, python made it possible to hardware optimize with libraries like numpy quite easily. Avoiding it is a mistake. I'll try to see if I have time to play the game myself and throw my attempt in there.

All the projects you listed are community projects that have grown over the years to become the backbone of scientific computing.

conda-forge, has evolved as one of the major conda community projects that helps release the latest releases from the projects you listed.

You can help improve the state of pytorch packaging on conda-forge too!

We've even released Pytorch 1.13 + Python 3.11 on linux and OSX! Give it a shot and let us know what you think!

edit: Link to the conda-forge pytorch development repository https://github.com/conda-forge/pytorch-cpu-feedstock

I was trying to find this thread again, https://jumpshare.com/

The logos look really similar....

I'll take a survey tomorrow to see how many people know what kelvin to C conversion is off the top of their head.

The other issue is when you get to Candelas!

No. Typically SI is implicit. Everything else is explicit.

I typically stick to "SI" units.

Then, somebody asks me to code in the temperature of a system. And I have to think: "Is now really the time that I want to teach people the difference a kelvins and celsius?"

So my rule becomes, SI "except" temperature. Sigh...

Ramona Optics | Embedded Systems Engineer | Durham, NC | Onsite

At Ramona Optics, we’re combining machine learning with novel optics to reinvent what’s possible with a microscope. Founded by Caltech graduates, Roarke Horstmeyer and Mark Harfouche, Ramona Optics is developing the first gigapixel microscope that can capture cellular-level detail over an entire square foot.

We are looking for an Embedded Systems Engineer to help bring Ramona Optics’ next generation of smart gigapixel microscopes to life. The person in this role will work closely with company founders to push the limits of the Ramona’s existing technology and help the business grow and evolve as we continue to develop our products and enter new markets. Your role will be to create the bridge between the information acquisition hardware, and the high-level analysis software. You will also be intimately involved in high-level decisions that company utilizes the hardware to manage the large volumes of acquired data -- 10s of gigabytes per second!

Email: hiring@ramonaoptics.com for more information

Thanks for the link.

I wonder how they are doing tail call optimization in python

And Coconut code runs the same on any Python version, making the Python 2/3 split a thing of the past.

My understand was that the big issue was the fact that [third party] libraries assume that strings were made of 8bits characters.

How does coconut solve that issue?

Seems a little misleading....

if you import enough things from `six` or `__future__` then you code will run the same on both python 2/3.

some things like fancy tuple expansion aren't available if you work with that subset, but yeah, most things are OK.

Yes.

conda basically proves that you can install almost everything one needs in the user's home directory. they have been working more and more on being completely independent from things like system compilers as well.

Often scientists are using hardware to acquire new data. The acquisition hardware might be on a PC that came installed from the manufacturer where you are told not to change anything.

Touching that PC, in anyway would be considered harmful to everybody using that specific piece of equipment.

Therefore, from the beginning of your acquisition, you are basically using a machine you don't control.

It is quite navigable. You can click anywhere on the table of content and it will get you right there. There is also a sidebar with bookmarks that work. You can basically skip 70 pages. Considering it is already 2000 pages, you have to use the table of contents already.

The scope of the PDF is to be the full documentation. If you were used to doing something a certain way, you need to know that things changed.

You can bring it up as an issue on github. They can probably move it down to the bottom.

The change log has A LOT of value if you are trying to figure out what version you should upgrade to. If you are stuck on an older version, then you need to understand what value you gain by potentially introducing new bugs into your product.

They also credit everybody who has worked on the project, which is super important.

When apps send me notifications on my phone, it is a reminder for me to uninstall them.

When sites send me annoying ~~adds~~ notifications, it is a reminder for me to turn on adblock for them.

edit: adds to notifications (also I can't spell obviously)

When I replaced an iPhone5c screen, the notion of selling the glass+LCD was still new. We used a hot gun to delaminate the cracked screen from the glass, then UV cured epoxy to glue the screen back (don't ask me why we had a UV gun). That was a pain.

Now, the glass + digitizer + LCD (or OLED) are sold as one piece. Sure you are "paying for a LCD/OLED" that "isn't broken", but the cost still comes down under $100 (typically) and saves probably 2 hours of labor.

I think the notion that "An apple product cannot be repaired" is also an internet spread cliche.

I found that it was pretty easy to replace the battery and screen of the last few cellphones I've owned. Ironically, it was much easier (and cheaper) to find the necessary parts to repair the apple ones than the cheapo android phones which often cost as much as the aftermarket parts.

If you consider the cost of specialized labour needed to diagnose the problem, then repairing quickly becomes unaffordable for a $300-$1000 purchase. This bill will not stop that.

Appliances are physically larger, and therefore are much more difficult to dispose of, ship, and get into place. So if you compare the cost of repair, to the shipping and disposal cost, they probably become equivalent. Goodluck getting a fridge out of your house without damaging the walls for cheap.