HP optimization levels and features

This section provides an overview of optimization features which can be through either the command-line optimization options or manual specification using pragmas or directives.

Five optimization levels are available for use with the HP compiler: +O0 (the default), +O1, +O2, +O3, and +O4. These options have identical names and perform identical optimizations, regardless of which compiler you are using. They can also be specified on the compiler command line in conjunction with other options you may want to use. HP compiler optimization levels are described in Table 3-2 “Optimization levels and features”.

Table 3-2 Optimization levels and features

Optimization Levels	Features	Benefits
`+O0` (the default)	Occurs at the machine-instruction level Constant folding Data alignment on natural boundaries Partial evaluation of test conditions Registers (simple allocation)	Compiles fastest.
`+O1` includes all of `+O0`	Occurs at the block level Branch optim ization Dead code elimination Instruction scheduler Peephole optimizations Registers (faster allocation)	Produces faster programs than `+O0`, and compiles faster than level `+O2`.
`+O2 (-O)` includes all of `+O0`, `+O1`	Occurs at the routine level Common subexpression elimination Constant folding (advanced) and propagation Loop-invariant code motion Loop unrolling Registers (global allocation) Register reassociation Software pipelining Store/copy optimization Strength reduction of induction variables and constants Unused definition elimination	Can produce faster run-time code than `+O1` if loops are used extensively.Run-times for loop-oriented floating-point intensive applications may be reduced up to 90 per cent.Operating system and interactive applications that use the optimized system libraries may achieve 30 per cent to 50 per cent additional improvement.
`+O3` includes all of `+O0,+O1,+O2`	Occurs at the file level Cloning within a single source file Data localization Automatic and directive-specified loop parallelization Directive-specified region parallelization Directive-specified task parallelization	Can produce faster run-time code than `+O2` on code that frequently calls small functions, or if loops are extensively used. Links faster than `+O4`.
	Inlining within a single source file Loop blocking Loop distribution Loop fusion Loop interchange Loop reordering - preventing Loop unroll and jam Parallelization Paralleliz ation, preventing Reductions Test promotion All of the directives and pragmas of the HP parallel programming model are available in the Fortran 90 and C compilers. `prefer_parallel` requests parallelization of the following loop `loop_parallel` forces parallelization on the last loop `parallel`, `end_parallel` parallelizes a single code region to run on multiple threads. `begin_tasks, next_task, end_tasks` forces parallelization of following code section
`+O4` includes all of `+O0, +O1`, `+O2`, `+O3` Not available in Fortran 90	Occurs at the cross-module level and performed at link time Cloning across multiple source files Global/static variable optimizations Inlining across multiple source files	Produces faster run-time code than `when +O3` global variables are used or when procedure calls are inlined across modules.

Cumulative Options

The optimization options that control an optimization level are cumulative so that 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 3-2 “Optimization levels and features”:

% 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 HP parallel programming model .

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 9 “Parallel programming techniques”.

HP optimization levels and features

Technical documentation

» Table of Contents

» Glossary

Cumulative Options