It is still a 64-bit OS code, but using 64-bit integer arguments for all functions.int64 refers to the default type of integer arguments expected by the FORTRAN library functions (all ACML functions are FORTRAN callable).
Consider the DSCAL routine as an example, which has the fortran interface: SUBROUTINE DSCAL (N, A, X, INCX)Here, A is the double precision scale factor and X is the double precision array. In this case the program would require that the integer parameters are 64-bits, and the int64 version of ACML would be used to supply the DSCAL routine.
Most applications will use the standard build (not int64). The user will often know if this is the case for gfortran because –fdefault-integer-8 is on the compile line for their fortran code.
Note that there is no performance penalty for using the int64 version.
KB18 This article explains the difference between the ACML int64 and 64bit versions.
Details and Recommendations Some of the 64bit ACML libraries provide an additional int64 version.
For instance there are two 64-bit gfortran downloads: acml-4-2-0-gfortran-64and acml-4-2-0-gfortran-64bit-int64So what’s the difference, and when should the int64 version be used?Both of the named tgz files are 64-bit versions of ACML.64-bit refers to the compiled code base: it’s compiled for 64-bit OS and 64-bit executable binary interface rules.But acml-4-2-0-gfortran-64bit expects 32-bit integer values when integers are part of a function’s argument list.The other library, acml-4-2-0-gfortran-64bit-int64 is the 64-bit, int64 library.