SciDAC QLA C interface Version 1.1
February 4, 2003

This is a second beta release of the QLA C library.  Like version 1.0
it has passed a battery of tests, but should be used with caution in
production running until we have had more experience with it.

Code generation is done through Perl scripts.  The most
straightforward workable code is produced with no particular effort
made to help the compiler with optimization.

Future releases will feature selected performance tests.

The changes from version 1.0 are outlined below.

Bugs and suggestions:
C. DeTar
University of Utah
detar@physics.utah.edu


----------------------------------------------------------------------
Changes from version 1.0

1. Dropped the equivalent Gauge and StaggeredPropagator types and
   consolidated to a "ColorMatrix" type with type code "M" (which
   was previously used for StaggeredPropagator).

2. Changed StaggeredFermion to ColorVector.

3. Introduced enhanced precision for the result of global reductions.

   A new long double type with code "Q" has been introduced.
   Its use is limited to returning results from global reductions
   and to type conversion back to double.

   Global reductions may now produce results in the same precision
   as the source, or in the next higher precision.

   e.g. QLA_QD3_m_eq_sum_M gives result as type QLA_Q3_ColorMatrix

   The addition of long double results requires the possibility of
   conversion from long double back to double.  e.g. QLA_DQ3_M_veq_M

   This change adds new libraries and headers with identifiers qla_dq,
   qla_dq3, qla_dq2, qla_dqn, qla_q, qla_q2, qla_q3, qla_qn, and new
   accessors, e.g. QLA_Q_elem_M.

4. New elementary unary functions:

    ceil, floor, cosh, sinh, tanh, log10

5. New elementary binary functions added to mod, min, max
  
    ldexp, pow, atan2

   The naming conventions may seem a bit idiosyncratic...

    QLA_F_R_eq_R_atan2_R, QLA_F_R_eq_R_pow_R, QLA_F_R_eq_R_ldexp_I
    

6. Local squared norm and local inner product

   Rather than performing a reduction, these operations generate
   a Real or Complex array.  e.g.

    QLA_F3_R_xeq_norm2_M, QLA_D2_C_veq_D_dot_D

   We also include the real part of the inner product, as in

    QLA_D3_R_veq_re_V_dot_V

   The allowed source types now include matrices as well as vectors,
   as illustrated.  The meaning of inner product for matrices A, B is
   the natural choice: Tr (A^\dagger B)

   We have tried to maintain the parallelism in reduction operations:

     QLA_D3_r_veq_re_D_dot_D

   The change to lowercase r distinguishes global from local
   operation.

7. Real to integer conversion...

   We now distinguish truncation and rounding.  Previously we had
   only truncation, so we change the name

      QLA_F_I_eqop_R -> QLA_F_I_eqop_trunc_R

   and introduce

      QLA_F_I_eqop_round_R

   Since "round" was first introduced in the standard C library in C99,
   we provide a version of "round" for older compilers.  The source
   code is in the subdirectory c99-src and the library is libqla_c99.a
   (just one module for now!).


8. Other added functions

    M_eqop_spintrace_P
    T_eqop_i_T (multiplication by i for all complex types)
    C_eq_C_divide_C
    I_eq_I_divide_I   


9. Integer reductions now produce only a Real type result:

    i_eqop_sum_I -> r_eqop_sum_I
    i_eqop_I_dot_I -> r_eqop_I_dot_I
    i_eqop_norm2_I -> r_eqop_norm2_I

10. A new real part of the global inner product

    e.g. QLA_F3_r_veq_re_D_dot_D

11. Miscellaneous name changes

      plus_i_times_R -> plus_i_R
      realtrace_M -> re_trace_M
      imagtrace_M -> im_trace_M
      Re -> re
      Im -> im

12. New complex macros for computing the real and imaginary parts
    of products:

      e.g. QLA_r_peq_Re_ca_times_c for Re(a^* * b)