HP UX Fortran Software manual +Oregreassoc, +Onoreturn, +Oshortdata=8, +Ovectorize option on

Page 98

Table 29 Fine-tuning optimization options (continued)

Level

Level

Function

+O[no]regreassoc

+O2or higher +O2or higher

Enable [disable] register

 

 

association. The default is

 

 

+Oregreassoc.

 

 

NOTE: This option is valid

 

 

only on the PA-RISC systems.

+Ono_return

All

This option provides a hint

 

 

to the optimizer that the

 

 

named functions never return

 

 

to the call site. If the

 

 

+Ono_return:filenameform

 

 

is used, the file filename

 

 

contains a list of function

 

 

names separated by spaces

 

 

or newlines. This

 

 

optimization is available at

 

 

all optimization levels.

+Oshortdata[=size]

All

All objects of size bytes or

 

 

smaller are placed in the

 

 

short data area, and

 

 

references to such data

 

 

assume it resides in the short

 

 

data area. Valid values of

 

 

size are a decimal number

 

 

between 8 and 4,194,304

 

 

(4MB). If no size is

 

 

specified, all data is placed

 

 

in the short data area.

 

 

The default is

 

 

+Oshortdata=8.

+[no]srcpos

All

This option controls the

 

 

generation of source position

 

 

information for HP Caliper.

 

 

The default is +srcpos.

 

 

When +srcpos, is in effect,

 

 

the compiler generates

 

 

source position information.

 

 

When+nosrcposis in

 

 

effect, the compiler does not

 

 

generate this information

 

 

and the compiler instructs the

 

 

linker to discard any of this

 

 

information found in the

 

 

object files.

+O[no]store_ordering

All

This option preserves [does

 

 

not preserve] the original

 

 

program order for stores to

 

 

memory that is visible to

 

 

multiple threads. This does

 

 

not imply strong ordering.

 

 

The default is

 

 

+Onostore_ordering.

+O[no]vectorize

+O3or higher

Replace [do not replace]

 

 

eligible loops with calls to

 

 

the math library; for more

 

 

information, see “Using the

 

 

+Ovectorizeoption” on

 

 

page168. The default

 

 

is+Onovectorize.

98 Performance and optimization

Image 98
Contents HP Fortran Programmer Guide AbstractPage Contents Debugging Using the on statementControlling data storage Performance and optimizationCalling C routines from HP Fortran 110 Using Fortran directives 123Writing HP-UX applications 107 Migrating to HP Fortran 131Documentation Feedback 153 Glossary 154 Index 159 Fortran 2003 Features 151Porting to HP Fortran 141 HP secure development lifecycle An overview of HP Fortran An overview of HP FortranHP Fortran compiler environment +dryrun DriverOptions for controlling the f90 driver +preinclude= filePreprocessor Options for controlling the C preprocessorFront-end Options for controlling the front end+moddir=directory Back-end Options for controlling optimizationOptimization Options for controlling code generation+Onooptimization +DAmodelLinker Options for controlling the LinkerOoutfile +FPflagsLdirectory Wl ,options ToolsHP-UX operating system Compiling and linking Compiling with the f90 commandF90 command syntax $ f90 hello.f90Command-line options Command-line optionsF90 command syntax Example 2 hello.f90Commonly-used options Command-line options by categoryCommonly-used options +saveOption descriptions Options listed by category+allowunaligned Example 3 ExampleDo I+1, N +autodbl +autodbl4 Data type sizes and +autodbl414164 Boption+check=bounds +cpp=default+charlit77 +nocfc+DDdatamodel Name=def+DAmodel DatamodelareItanium2 BlendedItanium NativeValues for the +FP option +hugecommon Signals recognized by the +fpexception optionGformat77 Example 4 % f90 +hugecommon=results pcvals.f90 +initheapcomplex=rvalival /usr/include directory +noimplicitnone+indirectcommonlist=file +initheapinteger=ival+nocheckuf +io77Ipo +nolibsWith different values of optlevel Levels of optimizationRequires concurrent use of the +Oprofile=use option +noobjdebug+pa1 +r8 +demandload option. The default is +nodemandload+nodemandload the default +realconstant=singleF90com Tx,pathTp,/usr/ccs/lbin/cpp End.oWx,arg1,arg2,...,argN Bextern =symbol ,symbol Symbol binding optionsBdefault=symbol,symbol Bhidden =symbol ,symbolF90 +O3 +Osize myprog.f90 Using optimization optionsReviewing general optimization options +Onoautopar +Oconservative+Onoall +Oautopar and omit +OparallelFine-tuning optimization options F90 +O3 +Onomoveflops +Ofltacc myprog.f90+Ocachepadcommon option Default is +OnocxlimitedrangeDefault is +Odataprefetch +Onocxlimitedrange+Onoentrysched +Onofenvaccess+Onofastaccess +OnofailsafeOptimizations performed by +Onofltacc +Onoinlinefilename +Oinlinebudget=n +Oinlinebudget enables+Onoinline +Onoinline=function1,function2Values for the +Oinlinebudget option Millicode versions of intrinsic functions+Oloopunroll=4 +inlinelevel num+Onoloopunroll=factor +Onoloopunrolljam+Onoparmsoverlap Default is+Onoparmsoverlap+Oparallelintrinsics +OnopipelineDefault is +Oshortdata=8 Default is +Onopromoteindirectcalls+Onorecovery For +Oprofile=collectarc,strideFilenames Filenames recognized by f90Linking HP Fortran programs Linking with f90 vs. ldLinking to libraries Libraries linked by default on PA-RISCLibraries linked by default on Itanium $ f90 -c hello.f90 # compileLinking to nondefault libraries Linking HP Fortran 90 routinesOpt/fortran90/lib/pa2064/ -lF90 -lisamstub Additional HP Fortran librariesLinking to shared libraries Library search rules Special-purpose compilationsCompiling programs with modules $ f90 -Wl,-a,archive prog.f90 -lmSpecial-purpose compilations Example 6 Example 2-2 main.f90 ExamplesExample Example 7 Example 2-3 code.f90$ f90 -o dostats data.f90 code.f90 main.f90 Compiling with makeExample 8 Example 2-4 data.f90 $ dostatsExample 9 Example 2-5 makefile Compiling for different PA-RISC machinesManaging .mod files $ makeCreating shared libraries Compiling with +picUsing the C preprocessor Linking with -bExample 13 Example 2-9 cppdirect.f90 Using the C preprocessorProcessing cpp directives $ f90 +cpp=yes -D Debug cppdirect.f90Saving the cpp output file Creating demand-loadable executablesCreating shared executables $ f90 +noshared prog.f90 Compiling in 64-bit modeUsing environment variables HP Fortran environment variablesHPF90OPTS environment variable F90ROOT environment variableSTF90COM64 environment variable $ f90 +list hello.f90Lpath environment variable Floating installationFloating installation Mpnumberofthreads environment variableSetting up floating installation Alternate-path/opt/fortran90.3.6.1Automatic and static variables Controlling data storageDisabling implicit typing Disabling implicit typingContains Controlling data storageIncreasing the precision of constants Increasing default data sizes Increasing default data sizesIncreasing default data sizes Which creates multiple threads Sharing data among programsUsr/lib/libpthread.sl $ gotosleep Sharing data among programsIm up Modules vs. common blocks$ wakeup Modules vs. common blocks Stripping debugging information Using the HP WDB debuggerDebugging Handling runtime exceptions Signals recognized by +fpexceptionSignal Floating-point exceptions= 1.0/0.0 Bus error exceptionFloating-point exceptions Illegal instruction exception Segmentation violation exceptionUsing debugging lines Bad argument exceptionOn REAL8 DIV 0 Call divzerotrap Using the on statementExceptions handled by the on statement Exceptions handled by the on statementExceptions handled by the on statement Actions specified by onOn Double Precision DIV 0 Call divzerotrap Example 14 Example5-1 abort.f90 Ignoring errorsTerminating program execution Example 15 Example5-2 ignore.f90Trapping integer overflow exceptions Calling a trap procedureTrapping floating-point exceptions On Double Precision Overflow Call trapExample 17 Example5-4 callitrap.f90 Trapping +Ctrl-C trap interruptsAllowing core dumps Example 18 Example 5-5 allowcore.f90 On Real Overflow IgnorePerformance and optimization Using profilersUsing profilers HP CaliperProgram.c Comparing Program PerformanceOpt/ansic/bin/cc -Aa +O3 -o program +Oprofile=collect ProgramprogramargumentsSpecifying PBO file names and locations Using Options to Control Data CollectionGprof $ gprof prog gprof.outProf Using options to control optimizationUsing +O to set optimization levels $ f90 +O4 file.f90+O3 Using the optimization options+O2, -O +O4$ f90 +O4 +Oaggressive +Ofltacc prog.f90 Fine-tuning optimization options$ f90 +02 +Oaggressive +Osize prog.f90 Packaged optimization options+O2 Is +Onofastaccess at+Ofastaccess at level +Ofltacc=relaxed+Onoinitcheck +Ofltacc=relaxed . ThisFast +Olibcalls +Oinlinelevel num+Onolibcalls +Onoloopunroll=n+Onoparminit +Opipeline+Orecovery +Oshortdata=8 +Oregreassoc+Onoreturn +Ovectorize option on+Owholeprogrammode Conservative vs. aggressive optimization+Onowholeprogrammode Compiling for parallel execution Conservative, aggressive, and default optimizationsParallelizing HP Fortran programs F90 +O3 +Oparallel -c x.f90 y.f90 F90 +O3 -c z.f90Conditions inhibiting loop parallelization Performance and parallelizationProfiling parallelized programs Calling routines with side effects parallellizationIndeterminate iteration counts Data dependencesF90 +O3 +Ovectorize prog.f90 Using the +Ovectorize optionVectorization Vector routines called by +OvectorizeSdot Controlling vectorization locallySaxpy VecdmultaddREAL, External sdot Calling Blas library routinesExample 19 Example 6-1 axpy.f90 Industry-wide standard VectorizationControlling code generation for performance $ fprog arg1 another arg Accessing command-line argumentsWriting HP-UX applications Example 20 Example 7-1 getargs.f90Performing I/O using HP-UX system calls Using HP-UX file I/OStream I/O using Fstream Calling HP-UX system and library routinesUsing HP-UX file I/O Obtaining an HP-UX file descriptorData type correspondence for HP Fortran and C Calling C routines from HP FortranData types Size differences between HP Fortran and C data types Unsigned integersLogicals Size differences after compiling with +autodblExample 21 Example 8-1 passcomplex.f90 Complex numbersComplex sqrcomplexCOMPLEX cmxval Pointers Argument-passing conventionsDerived types Example 22 Example 8-2 sqrcomplex.cVoid fooint *ptr, int iarray100, int Integer ptr INTEGER, DIMENSION100 iarrayCase sensitivity Call foo%REFptr, %REFiarray, %VALiExample 24 Example 8-4 testsort.f90 Example 23 Example 8-3 sortem.c$HP$ Alias bubblesort = BubbleSort%REF,%VAL Case sensitivityArrays Memory layout of a two-dimensional array in Fortran and CREAL, DIMENSION2,3,4 IntExample 25 Example 8-5 passarray.f90 Example 26 Example 8-6 getarray.cFortran hidden length argument StringsNull-terminated string Passing a stringFollowing are example C and Fortran programs StringsExample 28 Example 8-8 getstring.c File handlingExample 27 Example 8-7 passchars.f90 Example 29 Example 8-9 fnumtest.f90 File handlingExtern int somedata Sharing dataInt somedata Extern int globals100Directive syntax Using Fortran directivesUsing HP Fortran directives HP Fortran directives$HP$ Alias name = external-name arg-pass-mode-list SyntaxDescription and restrictions NameArgument-passing conventions Local and global usageCase sensitivity Example 31 Example 9-1 prstr.c StringsFor more information Example 32 Example 9-2 passstr.f90Example 33 Example Disables the inclusion of source lines in the listing fileSpecified on the command line Listing fileCompatibility directives recognized by HP Fortran Compatibility directivesControlling vectorization Vendor Directive CrayControlling checks for side effects Controlling parallelizationControlling dependence checks Compatibility directivesUsing Fortran directives Incompatibilities with HP Fortran Command-line options not supportedMigrating to HP Fortran Compiler limitsIntrinsic functions Format field widthsFloating-point constants Double Precision x =Procedure calls and definitions Data types and constantsFoo**REALbar, 8 ! foo**bar Input/outputDirectives KEY=Migration issues Migration issuesSource code issues MiscellaneousDirectives HP Fortran 77 directives supported by f90 optionsConflicting intrinsics and libU77 routine names Command-line option issuesIntrinsic functions F77 options supported by f90Object code issues Data file issuesApproaches to migration HP-supplied migration tools$ fid +800 file.f $ fid +es program.f Compatibility statements Porting to HP FortranCompatibility extensions END structure definitionPointer Cray-style Compiler directivesCompatibility directives +Oparallel orDirective prefixes recognized by HP Fortran Intrinsic proceduresNonstandard intrinsic procedures in HP Fortran +Oparallel or +OvectorizeUsing porting options Uninitialized variablesOne-trip do loops Using porting optionsLarge word size $ f90 testloop.f90External int1 Name conflictsExample 34 Example 11-1 clash.f90 Names with appended underscores Source formats+cfc Porting from Tru64 to HP FortranEscape sequences Nof66alternate for +noonetrip EnhancementsNew options Porting from Tru64 to HP FortranCheck noboundsoptions for example, -nocheckbounds +nopadsrc AltparamInput/output enhancements Fortran 2003 FeaturesInteroperability with C Miscellaneous enhancementsData enhancements Object orientation featuresFortran 2003 Features Documentation Feedback 153Glossary GlossarySo on. See also row-major order 155Also filename extension Memory fault 157See ttv 159 SymbolsIndex