Made drawing a couple of times faster

[ilyakooo0]

Stopped copying vectors around.

I don't have exact benchmarks, but after this PR clipText takes up half of the time spent on drawing, and clipText itself is not about three times faster.

[jtdaugherty]

Thanks for your patch!

I'm curious how you know that this is faster if you don't have benchmarks.

[ilyakooo0]

I had a specific application that I was trying to optimise. I have flamegraphs of the execution of the application.

They are kind of meaningless on their own.

I did a couple of iterations:

1. Find a function that takes up a large amount of the flamegraph
2. Optimise that part
3. Look at whether the optimisation significantly reduced the portion of the flamegraph that the function occupies

[ilyakooo0]

Plus, I just observed the subjective performance of using the app.

(I am constructing a large Image and scrolling it across the screen. 95% of time is spent drawing the Image. It went from ~1FPS to ~15FPS)

[ilyakooo0]

On second though, the flamegraphs might be useful if you look at something that I didn't optimise, since everything is scaled proportionately.

(These first two images are interactive SVGs, so you can download them and explore in more detail)

Before

After

---

If we look at the writeAttrSet call, which I didn't touch, then we can see how many times bigger it is in the "after" compared to the "before". (and the blaze.<> operator, though I am not sure where it is coming from)

Before

After

[jtdaugherty]

Thanks! I haven't seen one of these graphs before, so I'm not sure how to read it. Is the larger writeAttrSet an indication that something else took up less time? What is the X axis? (Also, the first two don't seem to be interactive for me.)

[ilyakooo0]

the larger writeAttrSet an indication that something else took up less time? What is the X axis?

The x axis is roughly the total execution time. And the horizontal bars show what portion of the execution time each function takes up. You can read more info in the "Summary" section here: https://www.brendangregg.com/flamegraphs.html

So, yes, if an unchanged function takes up more proportional time, it means that the other functions take up less time.

Also, the first two don't seem to be interactive for me.

It doesn't work in-browser for me either. When I download it and open in Gapplin, I can interact with it. I am not familiar with SVG implementation standards, so I don't really know where else you can open it for it to be interactive (:

[jtdaugherty]

This set of patches had fallen off of my radar so I started doing a bit of testing to refresh my memory. I tried this PR with tart and the behavior I see when trying to run the tool is:

tart: Data.Vector.Mutable: uninitialised element. If you are trying to compact a vector, use the 'Data.Vector.force' function to remove uninitialised elements from the underlying array.
CallStack (from HasCallStack):
  error, called at src/Data/Vector/Mutable.hs:215:17 in vctr-0.13.0.0-e19965ca:Data.Vector.Mutable

Interestingly, I do not see that exception on startup with matterhorn.

[jtdaugherty]

It's possible that exception is actually due to tart's own use of mutable vectors. I'll investigate and see whether that's the case.

[jtdaugherty]

Oh, I'm wrong - tart uses mutable arrays, not vectors.

[ilyakooo0]

It is very much possible that I missed some corner case.

It is now evident that merging this would require a comprehensive test suite.

Unfortunately, I no longer have time to invest into this PR ;(

Maybe we could pull out some obviously correct optimisations and merge only those?

In particular, I would suggest merging the small clipText optimisation.