 |
» |
|
|
|
|  | |
Previous releases of HP Fortran introduced these improvements: Command-line
flags | |
+multi_open—enables attaching multiple FORTRAN UNITs
to the same file by allowing the file to be opened in multiple Fortran
OPEN statements. +io77—suppresses the generation of the optional
leading "0" before the decimal point for real
numbers printed with the E and F edit descriptors. FORTRAN 77 suppressed
these leading zeros unless the NOSTANDARDIO flag was used. Along
with +gformat77, the +io77 flag provides formatted output compatibility with
the FORTRAN 77 product. +cat—concatenates all source files of the
same source form together, then compiles the concatenated source
all at once. This enables inlining at +O3 within the concatenated
file. -dynamic—used to generate a dynamically-bound
executable. +[no]es—Similar to +[no]extend_source except that character literals and hollerith constants
continued across a line boundary are not padded. This option provides compatibility
with FORTRAN 77's +es option. +getarg0 +getarg1—these options control the behavior of
the getarg intrinsic subroutine. +getarg0 requests the industry standard behavior for getarg, where an index value of zero causes the program
name to be returned. HP's FORTRAN 77 getarg intrinsic also implements this industry standard
convention. +getarg1 is used to request non-standard behavior, where
an index value of one causes the program name to be returned (older releases
of HP's Fortran behaved in this manner). The default is +getarg0. gformat77—requests the FORTRAN 77 style of formatting
a value of zero with the G edit descriptor. Fortran uses an F edit
descriptor when the value being written is zero, while FORTRAN 77
uses an E edit descriptor. +what—prints the what string for the Fortran
driver, providing version and patch numbers. +[no]signedzero—enables signed-zero support. This option
forces a floating point value of negative zero that appears as a
formatted output list item to be represented in the output record
with a leading "-". This option also changes the
behavior of the SIGN intrinsic. The default is +signedzero. +es—similar to the existing +extend_source option, but differs in that it does not pad a character
literal which spans a continuation line with blanks out to column
255. Instead, it behaves like FORTRAN 77's +es option, and only uses the actual blank characters
in the continued line in the source file. +es only affects fixed source form blank padding. +noes disables this behavior (the default is +noes, meaning no extended lines, and character literals
continued across lines are padded with blanks to column 72, the
same as FORTRAN 77's behavior). +fastallocatable—enables a different representation for
allocatable arrays in the object code produced by the compiler.
This alternate representation avoids problems in optimizing code
containing allocatable array references. Additionally, this alternate representation
for allocatable arrays is binary compatible with the old representation +noalign64bitpointers—disables correct alignment of pointers
in derived types when compiling for wide mode (+DA2.0W). Earlier
releases of HP Fortran improperly aligned such pointers, occasionally
leading to runtime aborts. Since this change introduces a potential
binary incompatibility, the +noalign64bitpointers flag is provided to maintain the old behavior.
Users who compile in wide mode (+DA2.0W)—and have derived
types that contain components with the POINTER attribute—should
recompile all source files that reference variables of that derived
type. Users who have successfully used such derived type variables
with older releases, and do not wish to recompile all affected source
files, should always specify +noalign64bitpointers when compiling affected source files. +Oopenmp—allows users to enable the OpenMP Directives.
+Onoopenmp will disable the OpenMP Directives.
+r8—changes 4-byte real constants, intrinsics,
and user variables to 8-byte reals (rather than the 4-byte default). +i8—changes 4-byte integer and logical constants,
intrinsics, and user variables to 8-byte integers (rather than the
4-byte default).
GETARG | |
The GETARG intrinsic subroutine has been modified to be compatible
with HP's FORTRAN 77 compiler and other vendors' implementations.
This change requires source code changes and recompilation of all
source files that call GETARG. The first argument to GETARG, the index of the command line argument desired,
is now zero based instead of one based. A value of one will now
cause GETARG to return the first command line argument after
the program name, instead of the program name, and a value of zero
now returns the program name, instead of blanks. Users wanting the old behavior can specify the following command
line options: +getarg1 Gather/scatter
prefetch directive | |
The C$DIR PREFETCH array directive was previously added to the HP Fortran
v2.3 compiler. This directive obtains prefetches for indirect array
references. You are responsible for ensuring that the loop bounds
are modified so that no out-of-bound array references are generated.
A clean-up loop is typically required to account for references
that cannot be prefetched. The syntax of this directive is:
C$DIR PREFETCH list-of-array-references Each array-reference must be syntactically and semantically
acceptable as a left-hand side of an ASSIGNment statement. Any user
array with elements identified in a PREFETCH directive does not
automatically generate prefetching. Below is an example of sample use of this directive: C$DIR PREFETCH A(J(I+10))
DO I=1,N-10
A(J(I))=B(I)+C(I)
ENDDO
DO I=N-9,N
A(J(I))=B(I)+C(I)
ENDDO In the above loop, automatic prefetching is still generated
for the B, C, and J arrays. For the A array, A(J(I+10)) will be
prefetched on the Ith iteration of the loop. Since J(I+10) may not
be defined for I+10 greater than N, you will have to manually split
off the last ten iterations of the loop. Fortran
95 features | |
A new parallel DO statement and construct was added to HP
Fortran. The form of this construct is: FORALL forall-triplet-spec-list body-of statements
END FORALL where: forall-triplet-spec-list is a list of the following: index-name=subscript:subscript[:stride] followed by an optional scalar mask expression. An example of a FORALL statement is shown below: REAL :: A(10,10), B(10,10) = 1.0 FORALL (I-1: 10, J=1:10, B(1,J) /=0. 0) A(I,J) = REAL(I+J-2) B(I,J) = A(I,J) +B(I,J)*REAL(I*J) END FORALL LibU77
routines provide Y2K compliance for HP-UX 11.0 | |
Two libU77 subroutines—datey2k and idatey2k—are now available in the Fortran compiler to
handle Year-2000 (Y2K) date-related issues on HP-UX 11.0. The +U77
flag must be issued with both subroutines. Although these are provided
for Y2K compliance, it is recommended that the standard date_and_time intrinsic be used instead of these functions. | | | |  | NOTE: The existing date and idate intrinsics do not have
a Y2K problem. They only return the last two digits of the year.
And they will continue to do so after the rollover. The behavior
of these intrinsics cannot be modified because existing code may
depend on the current behavior. | | | | |
datey2k is designed to replace the f90 date intrinsic.
Its function and arguments are the same as date's, except that the returned string contains a
4-digit year (mm-dd-yyyy) instead of a 2-digit year (mm-dd-yy). The syntax of the datey2k subroutine is as follows: subroutine datey2k(date)
character*11 date idatey2k is designed to replace the HP Fortran idate intrinsic. idatey2k returns the true year in its third argument, as
opposed to the idate intrinsic, which returns the number of years
since 1900 in its third argument. The syntax of the idatey2k subroutine is as follows: subroutine idatey2k(month, date, year)
integer month, day, year The libU77 routine idate has similar functionality to idatey2k
(it returns the true year), but its arguments are passed differently. In code where date is referenced,
replace date with datey2k. Also, make sure that datey2k's argument is at least 11 characters in length. In code where the idate intrinsic (not the libU77 idate routine) is used,
replace idate with idatey2k.
Additional
Improvements | |
Improved run time performance of Fortran
features, particularly allocatable arrays. +fastallocatable and +noalign64bitpointers command-line flags are discussed below in "Command-line
flags." A native, subset OpenMP implementation—This
new feature implements nine new directives, including PARALLEL,
PARALLEL DO, PARALLEL SECTIONS, DO, SECTION, SECTIONS, BARRIER,
CRITICAL, and ORDERED. More detailed information about this new
feature is supplied in Appendix B of the HP-UX Parallel Programming
Guide. Improved optimization flag defaults. Improved compiler support—The
compiler now supports the complete Fortran 95 standard.
DA1.1 Data
Prefetching — This feature is beneficial if you want to
run a single +Odataprefetch executable on both PA-RISC 1.1 and 2.0 architectures.
Previously, if you built a +Odataprefetch executable on a 2.0 machine, the 1.1 architecture
ignored it. The +DA1.1 +Odataprefetch flags allow programs to be built to run on both
the PA-RISC 1.1 and 2.0 architectures but with the same benefits
as 2.0 prefetching on 2.0 architectures. These flags cause DA2.0
prefetches to be inserted into the code in the same place they would
be for +DA2.0 +Odataprefetch. Sum Reduction Optimization — These
have been optimized to compute several partial sums to more efficiently
use PA-RISC 2.0 architecture, resulting in 2x runtime improvements.
This optimization is only available at +O2 or higher optimization
levels. V2600 support — Support
for V2600 server cache information has been added (HP-UX 11.0/11i
only). New Directive—The
MIXED_FORMATS directive, which exists in HP FORTRAN 77, now works
in HP Fortran v2.3 and above. This directive turns on or off mixed
formats in formatted input/output (I/O) statements. It allows a
numeric edit descriptor of a type different from that of the corresponding
I/O list item to be read from or written to.
The
syntax of the MIXED_FORMATS directive is: $HP$MIXED_FORMATS [ON|OFF] where
ON turns mixed formats on in formatted I/O statements, and OFF turns
mixed formats off in formatted I/O statements. Build-time improvements at
optimization level +O3.
|
Printable version
