CS 641 - Advanced Computer Architecture

Announcements

• Final exam and overall course letter grades are posted on NetRun. Lots of good work on the projects!
• The final exam was Tuesday at 10:15am in our usual classroom, Chapman 104. Also, turn in your project 3 final code on Blackboard by Tuesday, May 10 at midnight. You should have clean, working, tuned parallel code *and* a detailed performance analysis including speedup. Feel free to turn in the performance analysis in text as code comments, as a spreadsheet, or as a separate document.
• Be ready to briefly discuss your project 3 progress in class on Thursday. I posted your project 3 comments in the grading section on NetRun.
• In class on Tuesday, April 19, be ready to present a few minutes of details for your own project 3 topic, and be ready to provide some suggestions to other students as they describe their projects. In particular, at this early stage, the theoretical performance modeling work we've been discussing is extremely useful: problems with load imbalance, alpha beta, and Amdahl are predictable, and easiest to fix very early on!
  • Matt: MPI explicit load balancing, probably for Mandelbrot rendering.
  • Adam: Experimental validation of Amdahl's law.
  • Ben: Transaction library for multi-threaded hardware.
  • Donald: Network sockets measurement, tune, and time.
  • Lucas: Network visualizers in Python; convert to CUDA (or GLSL?) for speed.
  • Mike: MPI: batch convert .wav to .mp3 (or flac).
  • Ruth: Run Length Encoding on the GPU in CUDA: check GPU performance.
  • Vikram: Current neural network for checkers evaluations is currently 15k evals/second; wants to speed it up using OpenMP, SSE, and the GPU.
  • Bob: Divide & conquer problem written using: OpenMP, MPI, and possibly SSE.
  Protocol: After the presentations, I'd like you to turn in one text file with brief written helpful suggestions/comments for each presentation here on Blackboard. Turn in your comments before class on Thursday. I'd like to collect O(n^2) total pieces of useful advice via this process, although O(n*k) for k>=4 is acceptable.
• HW3 is posted on NetRun, and (now) due Tuesday, April 26. Coverage is CUDA, OpenMP, and AVX. Combining OpenMP and AVX, a 2011 CPU can actually approach 2008 GPU speeds!
• Turn in your Project 2 final draft code and performance analysis on Blackboard by midnight Thursday, March 31. The performance analysis writeup can be as simple as a README text file, or as complex as a PDF with graphs and pagenumbers: I'm mostly interested in seeing a clear analysis of where you think the bottlenecks are. Be sure to cite all your sources: turning in code you copied off the net without citation will be considered an honor code violation!
• Project 2 rough draft comments are posted on NetRun. The final draft, including benchmarks, is due Thursday! If you need some good timing code, you can use my NetRun timing routines like "time_function" and "print_time" on your own machine using "Download this file as a .tar archive" (the timing code is in lib/inc.c).
• Project 2 rough drafts are due Thursday, March 24 at midnight on Blackboard.
• Overall course grades are posted on NetRun.
• The midterm exam was Thursday, May 10th in class.
• Project 2 topics were due Thursday, March 3 in class.
• HW2 was on NetRun
• HW1, to build a working simulated CPU in Logisim, was due Tuesday, February 8 at midnight. Your CPU must include multiple registers, multiple arithmetic operations, and a reasonably complete instruction set. Turn in the .circ file and a screenshot on Blackboard
• Project 1 presentations were the first week of February:
• Tuesday, February 1:
• Pipelining, by Adam
• Branch prediction, by Lucas
• Out of order execution, by Vikram
• South bridge hardware, by Ben
• Probabilistic processors, by Ruth
  
• Thursday, February 3:
• Hyperthreading, by Bob
• Multicore's present and future, by Mike
• OpenFlow networks, by Donald
• The Accelerated Processing Unit (APU), by Matt
  
• Tuesday, February 8:
  • ARM processors, by Mariya

Lecture Notes

• 05_05_review, Course Review for Final Exam modified 2011-05-05
• 05_03_mandelbrot_MPI_EPGPU, Mandelbrot: in MPI and EPGPU modified 2011-05-04
• 04_28_PUP, Serializing and Deserializing with PUP modified 2011-04-28
• 04_28_MPI, The Message Passing Interface (MPI) modified 2011-04-28
• 04_26_sockets, Network Sockets modified 2011-04-25
• 04_21_top500, TOP500 Supercomputers modified 2011-04-25
• 04_14_lockfree, Lock-Free Multicore Datastructures modified 2011-04-14
• 04_12_AVX, Intel's Advanced Vector Extensions (AVX) modified 2011-04-12
• 04_07_load_balance, Load Balancing in OpenMP modified 2011-04-06
• 04_05_modeling, Performance Modeling: Amdahl, AMAT, and Alpha-Beta modified 2011-04-01
• 03_29_cache, Cache Layout for Uni and Multiprocessors modified 2011-03-29
• 03_24_multicore_programming, Multicore Programming Models: Threads vs OpenMP modified 2011-03-29
• 03_22_modern_multicore, Modern Multicore: Intel Sandy Bridge modified 2011-03-21
• 03_08_review, Things to Know for the Midterm modified 2011-03-07
• 03_03_CUDA_PTX, CUDA PTX: GPU assembly language modified 2011-03-07
• 03_01_GPU, Graphics Processor Unit programming with CUDA modified 2011-03-01
• 02_24_sse_movemask, SSE Branching: Performance Comparison modified 2011-02-24
• 02_22_sse_branching, SIMD Branching modified 2011-02-22
• 02_17_floats, Floating-Point Numbers modified 2011-02-22
• 02_15_sse, SIMD: x86 SSE and ARM VFP modified 2011-02-14
• 02_10_assembly, Comparing Modern x86 and ARM Assembly Language modified 2011-02-10
• 02_08_simd, Modern Parallelism: Superscalar, SMP, SMT, and SIMD modified 2011-02-10
• 01_25_instruction_frequency, Experimental Instruction Frequencies modified 2011-01-27
• 01_20_cpu_intro, Building Simple CPUs modified 2011-01-19

Resources

• The course syllabus has dates to remember and the grading policy.
  
• Logisim is a good Java circuit editor and simulator.
• We'll use NetRun for several of the homeworks in this class.