Hacking on liboil
=================


New Implementations
-------------------

  New implementations using GCC inline assembly code should go in
  liboil/${arch}, or liboil/i386_amd64 if they compile on both i386
  and amd64 architectures.  New implementations that are written in
  POSIX C should go in liboil/c.  New implementations written with
  MMX/SSE/Altivec instrinsics should go in liboil/mmx, liboil/sse,
  or liboil/altivec.

  Implementations copied from other projects should generally be
  given a separate directory under liboil/, similar to liboil/motovec.

  Things to check for when writing new implementations:

    - handle the n==0 case correctly

    - handle various array alignments correctly if the vectorized
      code has trouble loading misaligned data.  Sometimes this is
      only a problem on certain CPUs.

    - unrolled loops and vectorized code needs to handle any extra
      array elements at the end of array.

    - if a class has strides, they must be followed.

  Things implementations can assume:

    - n will never be negative

    - alignment of individual array members.  For example, if the
      type is "u32", you can assume that pointers and strides are
      multiples of 4.

  In general, if your implementation is enabled on the current CPU
  and 'make check' passes, it's probably a clean implementation.

  Broken implementations (i.e., code that is a work-in-progress)
  are still allowed, as long as the broken code is wrapped in
  #ifdef ENABLE_BROKEN_IMPLS/#endif.


New Classes
-----------

  Reference implementations for new classes should go in liboil/ref.

  The naming of new classes is a tricky business.  The goal is to
  make the name short, easy to remember and type, but descriptive
  enough to differentiate it from alternatives.  This policy has
  not always been followed in the past, so don't follow that lead.

  Try to:
  
    - Use full names instead of abbreviations.  Some abbreviations
      however are common and acceptable, such as "diff", "avg", or
      "abs".

    - Use a name that makes sense independent of the application
      that you may be copying it from.

    - Use nouns instead of verbs (thus, "difference" instead of
      "subtract", or "sum" instead of "add").

  Class names are made up of a base part that describes what the
  function does, appended with modifiers.  Common modifiers are
  for the type ("_f64", "_u8"), or to indicate inaccuracies or
  limitations in implementations ("_i10", "_l10").

  _i10 stands for "inaccurate by a factor of 10", where the baseline
  accuracy is 2^-52 for doubles and 2^-24 (I think).  It's a decilog
  scale, so _i20 is a factor of 100, etc.  The baseline accuracy comes
  from the least expressible number greater than 1.0.

  _l10 stands for "something is limited to 10".  Sometimes this means
  input range, e.g., _l15 is (was?) used for some function class that
  could only handle input values in the range [-1<<14, 1<<14-1].
  Another class uses this to mean that only 10 of the input values
  can be non-zero (oil_idct8x8theora_l10).

  These modifiers are obviously not well-thought-out.  In order to
  be useful, applications need to be able to make predictions about
  accuracy based on value of n, input values, etc.  I also don't
  think that simply defining new classes is a maintainable solution.

  Use of underscores in the base part of the class name is arbitrary.
  This may change in the future.

  New classes should not use the modifier "_ns", since non-strided
  arrays are the default.  

  Parameters should generally follow the order:

    i1, is1, i2, is2, ..., d1, ds1, ..., s1, ss1, ..., n, m

  After you add a new class, it's necessary to run 'make update' in
  the liboil/ directory to regenerate some built headers.



Test Machines
-------------

  Machines that ds has access to:

    power5.infradead.org - powerpc POWER5
    bombadil.infradead.org - powerpc PPC970MP (G5)