     The linear system solver assembly code 'solv.s' used in the 'matrix'
     segment of the library must be replaced by the corresponding C code
     'solv.c' when the target platform is not an Intel based processor.

     For some reason the GNU-C optimization of solv.c code produced
     unnecessary saves and restores in the innermost loop of this
     linear system solver code. This severly degraded performance,
     so the assembly code 'solv.s' was introduced. I believe that
     this problem is unique to Intel patforms using the GNU-C
     compiler.
