Sun Studio 12: Fortran Library Reference

2.3 Non-Standard Fortran 95 Intrinsic Functions

The following functions are considered intrinsics by the f95 compiler, but are not part of the Fortran 95 standard.

2.3.1 Basic Linear Algebra Functions (BLAS)

When compiling with -xknown_lib=blas, the compiler will recognize calls to the following routines as intrinsics and will optimize for and link to the Sun Performance Library implementation. The compiler will ignore user-supplied versions of these routines.

Table 2–2 BLAS Intrinsics


Function	Description
`CAXPY` `DAXPY` `SAXPY` `ZAXPY`	Product of a scalar and a vector plus a vector
`CCOPY` `DCOPY` `SCOPY` `ZCOPY`	Copy a vector
`CDOTC` `CDOTU` `DDOT` `SDOT` `ZDOTC` `ZDOTU`	Dot product (inner product)
`CSCAL` `DSCAL` `SSCAL` `ZSCAL`	Scale a vector

See the Sun Performance Library User’s Guide for more information on these routines.

2.3.2 Interval Arithmetic Intrinsic Functions

The following table lists intrinsic functions that are recognized by the compiler when compiling for interval arithmetic (-xia). For details, see the Fortran 95 Interval Arithmetic Programming Reference.

`DINTERVAL`	`DIVIX`	`INF`	`INTERVAL`
`ISEMPTY`	`MAG`	`MID`	`MIG`
`NDIGITS`	`QINTERVAL`	`SINTERVAL`	`SUP`
`VDABS`	`VDACOS`	`VDASIN`	`VDATAN`
`VDATAN2`	`VDCEILING`	`VDCOS`	`VDCOSH`
`VDEXP`	`VDFLOOR`	`VDINF`	`VDINT`
`VDISEMPTY`	`VDLOG`	`VDLOG10`	`VDMAG`
`VDMID`	`VDMIG`	`VDMOD`	`VDNINT`
`VDSIGN`	`VDSIN`	`VDSINH`	`VDSQRT`
`VDSUP`	`VDTAN`	`VDTANH`	`VDWID`
`VQABS`	`VQCEILING`	`VQFLOOR`	`VQINF`
`VQINT`	`VQISEMPTY`	`VQMAG`	`VQMID`
`VQMIG`	`VQNINT`	`VQSUP`	`VQWID`
`VSABS`	`VSACOS`	`VSASIN`	`VSATAN`
`VSATAN2`	`VSCEILING`	`VSCOS`	`VSCOSH`
`VSEXP`	`VSFLOOR`	`VSINF`	`VSINT`
`VSISEMPTY`	`VSLOG`	`VSLOG10`	`VSMAG`
`VSMID`	`VSMIG`	`VSMOD`	`VSNINT`
`VSSIGN`	`VSSIN`	`VSSINH`	`VSSQRT`
`VSSUP`	`VSTAN`	`VSTANH`	`VSWID`
`WID`

2.3.3 Other Vendor Intrinsic Functions

The f95 compiler recognizes a variety of legacy intrinsic functions that were defined by Fortran compilers from other vendors, including Cray Research, Inc. These are obsolete and their use should be avoided.

Table 2–3 Intrinsic Functions From Cray CF90 and Other Compilers


Function	Arguments	Description
`CLOC`	([C=]c)	Obtains the address of a character object
`COMPL`	([I=]i)	Bit-by-bit complement of a word. Use NOT(i) instead
`COT`	([X=]x)	Generic cotangent. (Also: DCOT, QCOT)
`CSMG`	([I=]i,[J=]j,[K=]k)	Conditional Scalar Merge
`DSHIFTL`	([I=]i,[J=]j,[K=]k)	Double-object left shift of i and j by k bits
`DSHIFTR`	([I=]i,[J=]j,[K=]k)	Double-object right shift of i and j by k bits
`EQV`	([I=]i,[J=]j)	Logical equivalence. Use IOER(i,j) instead.
`FCD`	([I=]i,[J=]j)	Constructs a character pointer
`IBCHNG`	([I=]i, [POS=]j)	Generic function to change specified bit in a word.
`ISHA`	([I=]i, [SHIFT=]j)	Generic arithmetic shift
`ISHC`	([I=]i, [SHIFT=]j)	Generic circular shift
`ISHL`	([I=]i, [SHIFT=]j)	Generic left shift
`LEADZ`	([I=]i)	Counts number of leading 0 bits
`LENGTH`	([I=]i)	Returns the number of Cray words successfully transferred
`LOC`	([I=]i)	Returns the address of a variable (See 1.4.32 `loc`: Return the Address of an Object)
`NEQV`	([I=]i,[J=]j)	Logical non-equivalence. Use IOER(i,j) instead.
`POPCNT`	([I=]i)	Counts number of bits set to 1
`POPPAR`	([I=]i)	Computes bit popluation parity
`SHIFT`	([I=]i,[J=]j)	Shift left circular. Use ISHFT(i,j) or ISHFTC(i,j,k) instead.
`SHIFTA`	([I=]i,[J=]j)	Arithmetic shift with sign extension.
`SHIFTL`	([I=]i,[J=]j)	Shift left with zero fill. Use ISHFT(i,j) or ISHFTC(i,j,k) instead.
`SHIFTR`	([I=]i,[J=]j)	Shift right with zero fill. Use ISHFT(i,j) or ISHFTC(i,j,k) instead.
`TIMEF`	()	Returns elapsed time since the first call
`UNIT`	([I=]i)	Returns status of BUFFERIN or BUFFEROUT
`XOR`	([I=]i,[J=]j)	Logical exclusive OR. Use IOER(i,j) instead.

See also 2.3.4.2 Memory Functions for a list of VMS Fortran 77 intrinsics.

2.3.4 Other Extensions

The Fortran 95 compiler recognizes the following additional intrinsic functions:

2.3.4.1 MPI_SIZEOF

`MPI_SIZEOF(` `x``,` `size``,` `error``)` Returns the size in bytes of the machine representation of the given variable, `x`. If `x` is an array, it returns the size of the base element and not the size of the whole array
x	input; variable or array of arbitrary type
`size`	output; integer; size in bytes of `x`
`error`	output; integer; set to an error code if an error detected, zero otherwise

2.3.4.2 Memory Functions

Memory allocation, reallocation, and deallocation functions malloc(), realloc(), and free() are implemented as f95 intrinsics. See 1.4.35 malloc, malloc64, realloc, free: Allocate/Reallocate/Deallocate Memory for details.