For each project, I provide a number of questions and key points to be
answered/addressed in your reports. Use the given links as starting points.
Do some research and find more -good- references to deepen your knowledge
on the matter.


---------------------------

Expression templates and their use in linear algebra:
  - What are expression templates (ET)?
  - How are they implemented? (Give details and examples)
  - Which tools/libraries make use of them? (Give an general overview of these libraries)
  - Which linear algebra operations benefit from this technique? (again, give examples)
  - Which are the limitations of ETs?
  * Keywords: Expression Templates, lazy evaluation, memory bound, compute bound.
  * Links:
     * Expression Templates by Todd Veldhuizen: http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.43.248
     * Techniques for Scientific C++ by Todd Veldhuizen: www.cs.indiana.edu/pub/techreports/TR542.pdf

---------------------------

Smart Expressions Templates and their use in linear algebra:
  - What are expression templates (ET) / smart expression templates (SET)?
  - What is the implementation idea/technique behind them? (Give details and examples)
  - Which tools/libraries make use of them? (Give an general overview of these libraries)
  - Which linear algebra operations benefit from this technique? (again, give examples)
  - Which limitations of ETs do SETs overcome and by means of what techniques?
  * Keywords: Expression Templates, Smart Expression Templates, memory bound, compute bound.
  * Links:
     * Expression Templates by Todd Veldhuizen: http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.43.248
     * arxiv.org/pdf/1104.1729
 
---------------------------

Blitz / Blaze / Eigen / Armadillo
  - Introduction: What is it? Motivation for the project to start? Target optimizations / use cases?
  - API
     + Operands and Datatypes 
     + Classes of operations provided
     + Features (triangularity, symmetry, etc)
     + Examples
  - Which problems/operations are suitable - Which are not?
  - Technology behind it / which optimizations are applied?
  - Custom and/or own code? Relies on the BLAS library?
  - Comparison of performance results.
  * Keywords: Blitz/Blaze/Eigen/Armadillo, (expression) templates, BLAS.
  * Links: 
     * http://blitz.sourceforge.net/
     * https://code.google.com/p/blaze-lib/
     * http://eigen.tuxfamily.org/index.php?title=Main_Page
     * http://arma.sourceforge.net/

---------------------------

TCE: loop fusion and loop tiling
  - Intro: what is TCE?
  - Which problem do they address? Main ideas/optimizations to address the problem?
  - Discuss the details of some of these ideas. 
    Focus on minimization of operation count and memory requirements, and especially on
    the decision algorithms for the loop transformations.
  - Show some performance results.
  * Keywords: tensor contraction engine, loop fusion, loop tiling.
  * Links:
     * http://www.csc.lsu.edu/~gb/TCE/
     * http://www.csc.lsu.edu/~gb/TCE/Publications/SpaceTime-PLDI02.pdf

---------------------------

Spiral: superoptimization
  - Introduction to Spiral.
  - General overview of its architecture: phases, transformation rules, search, etc.
  - Focus and details on the generation of small kernels for permutations based on the superoptimization technique (see link below).
  - Performance results.
  * Keywords: Spiral, FFT, superoptimization, SIMD, intrinsics.
  * Links
     * http://spiral.net/
     * http://spiral.ece.cmu.edu:8080/pub-spiral/abstract.jsp?id=160

---------------------------

LLVM: 
  a) Code generation and auto-vectorization
  b) Intermediate Representation (IR) and target description (.td) files

  - Introduction to LLVM.
  - General overview of the LLVM framework/architecture: from program to machine code.
  - Focus on the specific module(s) for each topic.
    a) Code generation and autovectorization.
    b) Intermediate representation and the description of the target architecture in the td files.
  * Keywords: LLVM, vectorization, SIMD, intermediate representation, target description files.
  * Links:
     * http://llvm.org/
     * http://llvm.org/docs/
     * (a) http://llvm.org/docs/Vectorizers.html
     * (b) http://llvm.org/docs/TableGen/LangIntro.html

---------------------------

Peephole optimization
  - General introduction to code generation and compiler optimizations.
  - What is the idea behind peephole optimization?
  - Give examples of characteristic peephole (local) optimizations.
  - Automatic generation of peephole optimizers.
  * Keywords: Code generation, peephole optimizers.
  * Links:
     * Compilers: Principles, Techniques, and Tools (Chapter 8). Available in the CS library.
     * Using Peephole Optimization on Intermediate Code: http://dspace.ubvu.vu.nl/bitstream/handle/1871/2606/11047.pdf
     * Automatic Generation of Peephole Superoptimizers: http://theory.stanford.edu/~aiken/publications/papers/asplos06.pdf

---------------------------

Roofline model: memory-bound operations
  - What is the roofline model?
  - Goals of the roofline model?
  - Which architecture information/knowledge is required?
  - How to get that information? (manufacturer, empirical, tools, ...).
  - Concept of arithmetic intensity, compute-bound vs memory-bound, ...
  - Give and idea and examples of at least 3 optimizations to improve memory-bound code.
  * Keywords: roofline model, arithmetic intensity, memory- vs compute-bound, stream.
  * Links:
      * An Insightful Visual Performance Model for Multicore Architectures: http://www.eecs.berkeley.edu/~waterman/papers/roofline.pdf
      * http://www.spiral.net/software/roofline.html