[patch] Ext_data for reference implementation

[Jules Bergmann]

This patch updates Ext_data for the reference implementation as follows:

  - basic compile-time extdata access (direct access and copy access)
    goes into core/extdata.
  - advanced compile-time extdata access (flexible access) stays in
    opt/extdata.
  - run-time extdata access stays in opt/rt_extdata.

All the core bits using Ext_data (such as solvers, matvec, signal 
processing other than Fft) should work with core/extdata.

Since Fft uses Rt_ext_data, this patch adds a simplified workspace that 
uses Ext_data instead.  This does not query the backend to find out what 
format it supports, rather it just gives the data as is.  This requires 
the backend to support arbitrary stride, which the C-VSIPL BE now does.

This patch will move two files (I haven't moved them in svn yet to 
minimize the time between moving and committing):
  - move opt/block_copy into core
  - move core/fft/workspace into opt

This patch also
  - Splits Block_fill out of block_copy into its own header.

  - Renames the VISP_IMPL_CVSIP_HAS_{TYPE} macros to _HAVE_.  We
    were using both names (HAS and HAVE), but only defining one.

  - Defines macros VSIP_IMPL_PROVIDE_FFT_{FLOAT, DOUBLE, LONGDOUBLE} in
    configure.ac to indicate for which types FFT should be tested.

  - Fixes other minor bits with the ref-impl.

With this patch,

  * most tests pass with the optimized library using C-VSIP has a backend
     - we're getting miscompares for convolution, correlation, and fir
     - the solver tests don't pass because they've either not been
       checked in (QR, Cholesky), they're broken (LU), or C-VSIP doesn't
       implement them (SVD).

  * most of the ref-impl tests pass with the ref-impl library
     - the solver tests don't pass for reasons above.

  * nearly all tests pass with the optimized library using other backends
     - fir is failing with IPP ... need to look into that.

Stefan, can you review the Fft related changes in particular?

				thanks,
				-- Jules

-- 
Jules Bergmann
CodeSourcery
jules at codesourcery.com
(650) 331-3385 x705
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: ried.diff
URL: <http://sourcerytools.com/pipermail/vsipl++/attachments/20061206/89e94589/attachment.ksh>