PR3 final results!

Programming topics can be found here.
Post Reply
bikker
Posts: 67
Joined: Fri Sep 27, 2013 12:40 pm

PR3 final results!

Post by bikker » Mon Apr 21, 2014 5:38 pm

Hi all,

I have graded the final assignment for PR3. You'll find the sheet attached to this post.

Some remarks:

The pass rate is really high this year. I saw a lot of very good work, clearly taking into account the directions regarding the process. There are no less than 3 perfect grades (rounded) for the final assignment; by Eric Polman, Max Oomen and Jan de Graaf. Congrats!

That being said, a couple of remarks / advice:

- I noticed most of you used the RDTSC timer for all measurements. There are two issues with this: first of all, better tools are available, which would give you more detailed insight. And secondly, if you use 12-digit numbers all over your report, things are hard to read. :) This happened a lot.
- Some of the reports would really have benefited from some graphs.
- Not everyone timed for vaying workloads. In this case, that was quite important; without this, you can't catch the O(N^2) behavior of tank vs tank testing.
- The glow was obviously the main thing in this app. However, quite a few students pretty much did that and little else... I can see why, considering the 4000% or so improvement, but there was definitely more to be done, even to that glow. Some people made the glow 30x faster (nice), but others got 100x (awesome).
- SIMD: I asked you to show your skills. Nearly everyone did the mountains, and nothing else. I decided to award a 6 for that, if it was decently executed. If you did a sloppy job (e.g., lots of data retrieval in the SIMD'ed loop, making it slower than the original code) you scored a 5. Any sign of SIMD in other places got you a 10 for SIMD (some people did glow blending with SIMD, pretty cool!).
- In general, not a lot of you went to the bone on this one. I did not penalize it too much because I told you to do optimizations that matters (as indicated by measurements), but still... Sometimes I expected a bit more. If I see 'Normalize' after 'Length' in a loop, it's costing you points for low level optimization.

But, measurements were done, and I am happy about that (that was a problem in previous years). We'll really need to do a profiler session in class in block 4 though. :) You cannot profile Killzone with RDTSC, I'm afraid.

That's it, thanks for playing, and if you have remarks / complaints / dead wishes after seeing your grades, let me know. Please refrain from complaining about homework grades, just the final assignment grades please. ;)

See you tomorrow!

AND: if you are in tomorrow's class, please note that the lecture starts at 9.20. I'll send an e-mail about this in a minute.
Attachments
results 2014.xls
(43 KiB) Downloaded 435 times

User avatar
132409
Posts: 2
Joined: Mon Sep 30, 2013 12:00 pm
Contact:

Re: PR3 final results!

Post by 132409 » Mon Apr 21, 2014 5:56 pm

Image

Post Reply

Who is online

Users browsing this forum: No registered users and 1 guest