COMPUTER ORGANIZATION AND ARCHITECTURE DESIGNING FOR PERFORMANCE NINTH EDITION

A  variety  of  studies  have  been  made  to  analyze  the  behavior  of  HLL  programs.
Table  4.8,  discussed  in  Chapter  4,  includes  key  results  from  a  number  of  studies.
There is quite good agreement in the results of this mixture of languages and appli-
cations. Assignment statements predominate, suggesting that the simple movement
of  data  is  of  high  importance.  There  is  also  a  preponderance  of  conditional  state-
ments  (IF,  LOOP).  These  statements  are  implemented  in  machine  language  with
some sort of compare and branch instruction. This suggests that the sequence con-
trol mechanism of the instruction set is important.
These  results  are  instructive  to  the  machine  instruction  set  designer,  indicat-
ing which types of statements occur most often and therefore should be supported
in an “optimal” fashion. However, these results do not reveal which statements use
the most time in the execution of a typical program. That is, we want to answer the
question:  Given  a  compiled  machine-language  program,  which  statements  in  the
source language cause the execution of the most machine-language instructions and
what is the execution time of these instructions?
To  get  at  this  underlying  phenomenon,  the  Patterson  programs  [PATT82a],
described  in  Appendix  4A,  were  compiled  on  the  VAX,  PDP-11,  and  Motorola
68000 to determine the average number of machine instructions and memory refer-
ences per statement type. The second and third columns in Table 15.2 show the rela-
tive  frequency  of  occurrence  of  various  HLL  statements  in  a  variety  of  programs;
the  data  were  obtained  by  observing  the  occurrences  in  running  programs  rather
than  just  the  number  of  times  that  statements  occur  in  the  source  code.  Hence
these  metrics  capture  dynamic  behavior.  To  obtain  the  data  in  columns  four  and
five (machine-instruction weighted), each value in the second and third columns is
multiplied by the number of machine instructions produced by the compiler. These
results  are  then  normalized  so  that  columns  four  and  five  show  the  relative  fre-
quency  of  occurrence,  weighted  by  the  number  of  machine  instructions  per  HLL

statement. Similarly, the sixth and seventh columns are obtained by multiplying the
frequency of occurrence of each statement type by the relative number of memory
references caused by each statement. The data in columns four through seven pro-
vide  surrogate  measures  of  the  actual  time  spent  executing  the  various  statement
types. The results suggest that the procedure call/return is the most time-consuming
operation in typical HLL programs.
The reader should be clear on the significance of Table 15.2. This table indi-
cates the relative performance impact of various statement types in an HLL, when
that HLL is compiled for a typical contemporary instruction set architecture. Some
other architecture could conceivably produce different results. However, this study
produces  results  that  are  representative  for  contemporary  complex  instruction  set
computer  (CISC)  architectures.  Thus,  they  can  provide  guidance  to  those  looking
for more efficient ways to support HLLs.