• Arithmetic exceptions are the same as, or extensions of, exceptions for the
scalar arithmetic instructions of the System/370, and similar fix-up routines
can be used. To accommodate this, a vector interruption index is employed
that indicates the location in a vector register that is affected by an exception
(e.g., overflow). Thus, when execution of the vector instruction resumes, the
proper place in a vector register is accessed.
• Vector data reside in virtual storage, with page faults being handled in a standard manner.
This level of integration provides a number of benefits. Existing operating systems can support the vector facility with minor extensions. Existing application programs, language compilers, and other software can be run unchanged. Software that
could take advantage of the vector facility can be modified as desired.
REGISTERSA key issue in the design of a vector facility is whether operands are
located in registers or memory. The IBM organization is referred to as register to
register,because the vector operands, both input and output, can be staged in vector registers. This approach is also used on the Cray supercomputer. An alternative approach, used on Control Data machines, is to obtain operands directly
from memory. The main disadvantage of the use of vector registers is that the
programmer or compiler must take them into account for good performance. For
example, suppose that the length of the vector registers is Kand the length of the
vectors to be processed is NK. In this case, a vector loop must be performed,
in which the operation is performed on Kelements at a time and the loop is repeated times. The main advantage of the vector register approach is that the
operation is decoupled from slower main memory and instead takes place primarily with registers.