Five
optimization levels are available for use with the Exemplar compilers.
These options have identical names and perform identical optimizations,
regardless of which compiler you are using. They are specified on
the compiler command line along with any other options you wish
to use. Exemplar compiler optimization levels are summarized in
Table 3-1 “Compiler
optimization levels”.
Table 3-1 Compiler
optimization levels
| Option | Description |
|---|
+O0
(default) | (Machine instruction-level optimizations)
Constant
folding and simple register assignment |
| +O1 | (Block-level optimizations)
+O0
optimizations, plus instruction scheduling and optimizations on
basic blocks (A basic block is a linear sequence of machine instructions
with a single entry and a single exit.) |
| +O2 | (Routine-level optimizations)
+O1
optimizations, plus optimizations within subprograms in a single
file; loop optimizations to reduce pipeline stalls; analysis of
data flow, memory usage, loops, and expressions |
| +O3 | (File-level optimizations)
+O2
optimizations, plus full optimizations across all subprograms (including
inlining) within a single file; use of parallelism-related directives
and pragmas from the Exemplar programming model when +Oparallel
is also specified |
| +O4[1] | (Cross-module optimizations)
+O3
optimizations, plus full optimizations across the entire application;
optimizations include inlining across the application; the +O4
optimizations are performed at link time |
These options are cumulative; each option retains the optimizations
of the previous option. For example, entering the following command
line compiles the Fortran program foo.f with all +O2,
+O1, and +O0
optimizations shown in Table 3-1 “Compiler
optimization levels”.
% f90 +O2 foo.f
In addition to these options, the
+Oparallel option is available
for use at +O3 and above. (+Onoparallel
is the default.) When the
+Oparallel option is specified,
the compiler:
Looks for opportunities for parallel
execution in loops.
Honors the parallelism-related directives and pragmas
of the Exemplar programming model. When using Exemplar Fortran 77 Version
1.2.3 or Exemplar C Version 1.2.3, +Oexemplar_model
(the default) must also be in effect
for these directives and pragmas to be enabled.
The +Onoautopar
(no automatic parallelization) option is
available for use with +Oparallel
at +O3 and above; +Oautopar
is the default. +Onoautopar causes
the compiler to parallelize only those loops that are immediately
preceded by loop_parallel or prefer_parallel
directives or pragmas; for more information, refer to Chapter 4,
"Chapter 4 “Basic shared-memory
programming”."
The +Onodepar (node-parallelism)
option is also available for use with
+Oparallel at +O3
and above. This option causes the compiler to generate node-parallel
code (indicated by directives and pragmas that use the nodes
attribute) for a multinode, scalable SMP. (See Chapter 4, "Chapter 4 “Basic shared-memory
programming”," for information
on attributes.)
The +Ononodepar option (the default) causes the compiler to generate
code for a single-node machine. When this option is used, serial
code is generated for node-parallel constructs; thus, node-parallelism
is not implemented. Thread-parallelism—both automatic and
directive-specified—is still implemented.
Printable version
