VMS Help FORTRAN, /MATH_LIBRARY  *Conan The Librarian (sorry for the slow response - running on an old VAX)

  /[NO]MATH_LIBRARY=(option[,...])        D=/MATH_LIBRARY=(ACCURATE,NOV5)

  Specifies the selection of math library routines used to implement
  mathematical intrinsic functions.  The qualifier is not negatable
  and a value must be specified.  If the ACCURATE and FAST keywords
  are both specified, the last keyword is used.

1 - ACCURATE

  Directs the compiler to select the standard version of the math
  library routines used to implement vectorized references to the
  ATAN, COS, EXP, LOG, LOG10, SIN, SQRT and TAN intrinsics as well as
  the exponentiation operator ** for real data types.  This version
  gives the most accurate result.  For certain ranges of input
  values, the selected routine can execute more slowly than the
  alternate version specified by FAST.  This keyword is not
  negatable.

2 - FAST

  Directs the compiler to select the Fast Vector Math version of
  library routines used to implement vectorized references to the
  ATAN, COS, EXP, LOG, LOG10, SIN, SQRT and TAN intrinsics as well as
  the exponentiation operator ** for real data types.  This version
  gives the highest execution performance.  For certain ranges of
  input values, the selected routine may not give a result that is as
  accurate as the version specified by ACCURATE.  Accuracy reduction
  will be at most 2 bits for single-precision routines, and at most 4
  bits for double-precision routines.  This keyword is not negatable.

3 - V5

  Controls whether the compiler restricts its use of math library
  routines to those provided in OpenVMS VAX V5.4.  If V5 is
  specified, the compiler does not use optimized versions of the math
  library routines for scalar references to the AMOD, ATAN, ATAN2,
  CABS, CCOS, CEXP, CLOG, COSH, CSIN, CSQRT, SIGN, SINH and TANH
  intrinsic functions nor does it vectorize references to the ATAN2,
  CCOS, CEXP, COSH, CSIN, CSQRT, MOD, SINH and TANH intrinsic
  functions.  The default is NOV5 which enables the above
  optimizations and vectorizations, making use of new math library
  routines which are provided with the Compaq Fortran 77 product but which
  are not present in OpenVMS VAX versions V5.4 through V6.0.