Final Work Schedule

Final Work Schedule

25/2: Martin:

• Cancel OpenCL course ✓
• Cancel Åbenthus (find replacement) ✓ Troels Henriksen has promised to be there
• Collect final benchmark data
• Create crude plots
• Add LSM as benchmark

Philip:

• Expressivity survey: Write example of work around lack of nesting and describe in the report (see Accelerate Sobol code) ✓
• Expressivity tasks: consider memory allocation, general recursion ✓
• Nikola: compilation of individual kernels and Haskell to glue them together. Compare with Accelerate.

26/2: Martin:

• Write about least square solvers in "Cases"
• Write about portfolio pricing in "Cases"

27/2:

28/2:

Martin: Brahms Klaverkoncert i koncerthuset

1/3: Philip:

• Write about our added nikola combinators

2/3: Weekend

3/3: Weekend

Extension implementation status:

• fold (iterate-based)

  • Nikola doesn't tread IterateE with a DelayedE inside consistently, with detectSharing sometimes rewriting to a LetE. Results in cannot-compile errors. Possibly fixable by reconstructing LamE-invariant inside compileExp.

• fold2,unfoldP,mapNest (hand-coded push arrays)

  • Nikola's handling of variables doesn't allow for cross-iteration data dependencies in sequential for-loops. Possibly fixable by switching to IterateE, but would be nice to fix properly.

• sequential

  • Currently implemented as SequentialExecE in S.Exp, and manifest in compilation through the CEnv state. The compileExp case for ForE then compiles parallel or sequential loops depending on ForLoop setting and whether the loop body turns sequential. Quite possibly unsafe and open to abuse.