HP UX Performance Tools 237

What you need to know to use these subprograms

Table 3-4shows the valid combinations of T, XX, and Y, YY, or YYY. Each line indicates the allowable T preﬁxes and Y, YY, or YYY sufﬁxes for a particular root name XX.

Table 3-4 Extended BLAS Naming Convention—Subprogram Names

	Valid T			XX			Valid Y, YY, or YYY
S	D			GE	MM	MV	R
		C	Z	GE	MM	MV				RC	RU
S	D	C	Z	GB		MV
		C	Z	HE	MM	MV	R	R2	RK	R2K
		C	Z	HB		MV
		C	Z	HP		MV	R	R2
S	D			SY	MM	MV	R	R2	RK	R2K
		C	Z	SY	MM				RK	R2K
S	D			SB		MV
S	D			SP		MV	R	R2
S	D	C	Z	TR	MM	MV				SM	SV
S	D	C	Z	TB		MV					SV
S	D	C	Z	TP		MV					SV

The subprograms SGEMMS, DGEMMS, CGEMMS, and ZGEMMS, although not part of the standard Extended BLAS, are consistent with this nomenclature.

Operator arguments in the BLAS Standard

Some routines in the BLAS Standard take input-only arguments called operators. Operators allow for the speciﬁcation of multiple related operations to be performed by a single function. The BLAS Standard speciﬁes the type and the valid values these arguments should have according to the speciﬁc programming language.

Operator arguments used by the BLAS Standard routines are NORM, SORT, SIDE, UPLO, TRANS, CONJ, DIAG, and JROT. Refer to “Operator arguments” on page 25 for explanations of the valid operator values.

In BLAS Standard routines, you specify an operator argument with a named constant value. The actual numeric value assigned to the named constant is deﬁned in the appropriate language’s include ﬁle. Operator arguments are represented in the Fortran 77 interface as INTEGERs. This speciﬁcation is different from the legacy BLAS, where operator arguments are deﬁned as CHARACTER*1.

Refer to individual routines in “BLAS Standard routines” on page 339 for the named constants you can use to specify operator arguments for basic matrix subprograms.

Chapter 3 Basic Matrix Operations 209