Skip to content
Commits on Dec 1, 2011
  1. Provide scalar and piecewise AND

    Steve Kerrison committed
  2. Provide shift left.

    Steve Kerrison committed
  3. Add arithmetic shift right

    Steve Kerrison committed
  4. Fix SHR op to be a true logical shift.

    Steve Kerrison committed
  5. Update module_build_info to use -O3

    Steve Kerrison committed
Commits on Nov 30, 2011
  1. Provide a shift-right scalar op

    Steve Kerrison committed
Commits on Nov 24, 2011
  1. Replace old matrix_mul_worker with matrix_mul_worker_new

    Steve Kerrison committed
    Plus some comment cleanup that I left into the previous commit for reference.
  2. Update XC demo to do a bunch of new performance tests.

    Steve Kerrison committed
    The by-hand matrix demo could probably benefit from a little more
    optimisation, but it's indicative of what somebody might use in
    a hurry.
  3. Increase paralellism threshold for matrix multiplication to avoid blo…

    Steve Kerrison committed
    …ck size errors
  4. Tidy up MATRIX_WORKER_SPAWN macro

    Steve Kerrison committed
  5. Introduce new matrix multiplication method

    Steve Kerrison committed
    This one actually works, and can be split into workers in a similar
    way to the scalar and array ops.
    
    Still need to assess performance.
Commits on Nov 23, 2011
  1. Increase size of matrix performance test.

    Steve Kerrison committed
    Demonstrate that matrix macros work for global variables too (and why wouldn't they?).
  2. Break apart sca/arr ops to give much better performance.

    Steve Kerrison committed
    They're too deep into the op loop to make it worth thinking about saving
    code size.
  3. Always produce a resource report for the XC app.

    Steve Kerrison committed
Commits on Nov 18, 2011
  1. Improve matrix multiplication performance.

    Steve Kerrison committed
  2. Improve baseline and library scalar op performance.

    Steve Kerrison committed
Commits on Nov 17, 2011
  1. Add random matrix generation

    Steve Kerrison committed
  2. Change default number of threads spawned

    Steve Kerrison committed
  3. Fix unsafe switch case before default

    Steve Kerrison committed
  4. Change the way pointers are passed by value between XC/C

    Steve Kerrison committed
  5. Fix XC app makefile to use correct thread count definition.

    Steve Kerrison committed
  6. Update XC app to show performance of library

    Steve Kerrison committed
    Useful to see when it's worth parallelising matrix ops.
    Will help guide optimisation efforts.
  7. Code size and header fixes

    Steve Kerrison committed
    Reduce code size in worker op selection
    Fix MATRIX_REDIM to correctly assess matrix limits.
    Move number of threads definition.
  8. Fix .gitignore

    Steve Kerrison committed
  9. Enable a set of scalar ops on matrices

    Steve Kerrison committed
    Uses same approach as array (piecewise) ops.
  10. Enable array (piecewise) operations

    Steve Kerrison committed
    Combines several arithmetic operations into a single worker function.
Commits on Oct 17, 2011
  1. Fix array multiplication (which did add instead)

    Steve Kerrison committed
    Add support for scalar multiplication.
    Fix treatment of nullable entities for in-place results.
    Extend demo apps to use in-place scalar multiplication.
    
    Signed-off-by: Steve Kerrison <steve@stevekerrison.com>
  2. Add support for array-multiplication

    Steve Kerrison committed
  3. Minor tweak to readme.

    Steve Kerrison committed
    Signed-off-by: Steve Kerrison <steve@stevekerrison.com>
  4. Update readme.

    Steve Kerrison committed
    Signed-off-by: Steve Kerrison <steve@stevekerrison.com>
  5. Initial fit-for-consumption version of matrix multiplication.

    Steve Kerrison committed
    Signed-off-by: Steve Kerrison <steve@stevekerrison.com>
Commits on Oct 13, 2011
  1. Initial commit

    Steve Kerrison committed
Something went wrong with that request. Please try again.