Technical data

Algorithm Optimization Examples

This appendix illustrates how the characteristics of the VAX 6000

series vector processor can be used to build optimized routines for this

system and how the algorithm and its implementation can change the

performance of an application on the VAX 6000 processor.

The VAX 6000 series vector processor delivers high performance for

computationally intensive applications. Based on CMOS technology, the

VAX 6000 Model 400 vector processor is capable of operating at peak

speeds of 90 MFLOPs single precision and 45 MFLOPs double precision.

Linear algebra and signal processing applications that utilize the

various hardware features have demonstrated vector speedups between

3 and 35 over the scalar VAX 6000 CPU times. With the integrated

vector processing available on the VAX 6000 series, the performance

of computationally intensive applications may now approach that of

supercomputers.

Algorithm changes can alter the data access patterns to more efﬁciently

use the memory subsystem, can increase the average vector length,

and can minimize the number of vector operations required. By

applying Amdahl’s Law of vectorization, performance can be improved

by increasing the percentage of code that is vectorized.

Four basic optimization methods that take advantage of the processing

power of VAX 6000 series system include:

• Rearrange code for maximum vectorization of the inner loop and

remove data dependencies within the loop

• Vectorize across contiguous memory locations to produce unity stride

vectors for increased cache hit rates and optimized cache miss

handling

• Reuse the data already loaded into the vector registers as frequently

as possible to reduce the number of vector load and store operations

• Maximize instruction execution overlap by pairing arithmetic

instructions between load and store instructions wherever possible

Further information on optimization techniques in FORTRAN can be

found in the VAX FORTRAN Performance Guide available with the

FORTRAN-High Performance Option.

A–1