Technical data

Vector Processing Concepts

for which vectorization will yield correct results and generate vector

instructions for those operations. If the compiler cannot be certain that

a particular expression can be correctly vectorized, it will not vectorize

that portion of the source code. The vectorizing compiler can reorganize

sections of the program code (usually inside formal loops) that can be

vectorized.

Certain portions of all applications are nonvectorizable. Some

programming techniques, by their nature, cannot be vectorized. For

example, conditional branches into or out of a loop make it impossible

for the compiler to know the range of the loop (that is, the vector length)

before the code is executed.

In other instances, there may be an unclear relationship between multiple

references to the same memory location (called an unknown dependency).

In such a relationship, the ﬁnal value of the location may or may not

depend on serial execution of the code, and the compiler does not have

enough information to determine whether it can vectorize.

Finally, there may be instances of constructs that could be vectorized

but are not. The compiler may not be sophisticated enough to do so, the

compiler may determine that vectorization would not be proﬁtable in

terms of performance, or the compiler may have insufﬁcient information

to determine that vectorization is safe.

1.4 VECTOR REGISTERS

A scalar register is a location in fast memory where data is stored or

status bits can be placed to be read at a later time. A register has a set

length, say 32 bits, or four consecutive 8-bit bytes.

A vector register is considerably larger. With the VAX, the vector register

has a maximum length of 64 elements. Each element can contain up to

64 bits. The elements used can be enabled or disabled by setting bits in

a Vector Mask Register (VMR). The programmer usually determines the

range, or limits the number of elements used through a Vector Length

to 64 elements. Of course, if the vector length = 0, no vector elements will

be processed.

1–9