By Fabiha Hannan on 25 February 2014

Here are the minutes from 2/25/14.

In Attendance: Prof Spjut, Donghyeon, Fabiha, and Akhil Start time: 12:05 pm End time: 1:01 pm

Updates:

  • Akhil finished running the instruction counter for all of the benchmarks. This blog post has been updated with the numbers.

  • I finished writing a parser to extract information and calculate the stalls due to the instruction cache. I couldn't work on the instruction duplication because we have not yet edited and looked at the banks.

  • Donghyeon edited the instruction cache output to display the stats for each core's performance and made a more thorough list of each benchmark and its characteristics. These are in these two blog posts respectively: Instruction Cache Output and Sphynx Benchmarks.

Things to do this week:

  1. Make a table of GPU architectures & specs (Akhil)

  2. Work on the parser (Fabiha)

  3. See if =decreasing cache size makes a difference (same block size, reduce blocks to 1 set for 4-way set associativity (or from 1024 bytes to 512 bytes)) in the misses (Donghyeon)

  4. Independent assignment for everyone:

a) Come up with designs for shared cache [note: instruction memory tends to have sequential pattern & temporal repetition (loop)]

b) Skim a micro paper from last year and get a feel of what it should look like (Micro 2013 paper microarch.org)

Other notes:

  • The default for instruction caches: instruction copied to each core and must merge 2 cores (aka put execution units together) to get it to share

  • We are submitting to the IEEE microarchitecture conference: middle to end of May, we need it done at the beginning of May to submit to peer review

category: tags:


blog comments powered by Disqus