-----------------------------------------------------------------------------------
Fujitsu PRIMEPOWER flags/tunables description (Dec.14 2005)
(Each section is sorted in case insensitive, alphabetical order)
Table of Contents
[1] Sun Studio 11 flag description
[2] Environment Variables
[3] Kernel Parameters (/etc/system)
[4] Commands for feedback control
-----------------------------------------------------------------------------------
[1] Sun Studio 11 flag description
Compiler options Remark
-----------------------------------------------------------------------------------
cc Invoke the Sun Studio 11 Compiler C
(C compiler)
CC Invoke the Sun Studio 11 Compiler C++
(C++ compiler)
-crit Enable optimization of critical control paths
(optimizer)
-dalign Assume data is naturally aligned.
(C, C++, Fortran)
-Dalloca=__builtin_alloca
(Portability flag) Portability switch, used for 176.gcc:
allow use of compiler's internal builtin alloca.
-depend Synonym for -xdepend.
(Fortran)
-DHOST_WORDS_BIG_ENDIAN Portability switch, used for 176.gcc:
(Portability flag) controls how bytes are numbered within a word.
-D__MATHERR_ERRNO_DONTCARE
(C) Allows the compiler to assume that your code
does not rely on setting of the errno variable.
-DSPEC_CPU2000_SOLARIS Portability switch, used for 253.perlbmk:
(Portability flag) selects header files and code paths compatible
with Solaris.
-DSUN Portability switch, used for 186.crafty:
(Portability flag) selects header files and code paths
compatible with Solaris.
-DSYS_HAS_CALLOC_PROTO Portability switch, used for 254.gap:
(Portability flag) allows use of the designated prototype.
-DSYS_HAS_IOCTL_PROTO Portability switch, used for 254.gap:
(Portability flag) allows use of the designated prototype.
-DSYS_HAS_SIGNAL_PROTO Portability switch, used for 254.gap:
(Portability flag) allows use of the designated prototype.
-DSYS_HAS_TIME_PROTO Portability switch, used for 254.gap:
(Portability flag) allows use of the designated prototype.
-DSYS_IS_USG Portability switch, used for 254.gap:
(Portability flag) selects code compatible with USG-based systems.
-e Portability switch, used for 178.galgel:
(Portability, Fortran) allows source lines to be up to 132 characters long.
f90 Invoke the Sun Studio 11 Compiler Fortran 90
(Fortran compiler)
-fast A convenience option, this switch selects the
(C) following switches that are defined elsewhere
in this page:
-D__MATHERR_ERRNO_DONTCARE
-fns
-fsimple=2
-fsingle
-xalias_level=basic
-xarch=generic
-xbuiltin=%all
-xcache=generic
-xchip=generic
-xdepend
-xlibmil
-xlibmopt
-xmemalign=8s
-xO5
-xprefetch=auto,explicit
-xtarget=native
-fast A convenience option, this switch selects the
(C++) following switches that are defined elsewhere
in this page:
-fns
-fsimple=2
-xbuiltin=%all
-xlibmil
-xlibmopt
-xO5
-xtarget=native
-fast A convenience option, this switch selects the
(Fortran) following switches that are defined elsewhere
in this page:
-dalign
-fns
-fround=nearest
-fsimple=2
-ftrap=common
-xlibmil
-xlibmopt
-xO5
-xpad=local
-xprefetch=auto,explicit
-xtarget=native
-xvector=yes
-fixed Portability switch, used for 178.galgel:
(Portability, Fortran) assume fixed-format source input.
-fns Selects faster (but nonstandard) handling of
(C, C++, Fortran) floating point arithmetic exceptions and
gradual underflow.
-fsimple=<n> Controls simplifying assumptions for
(C, C++, Fortran) floating point arithmetic:
-fsimple=0 Permits no simplifying assumptions.
Preserves strict IEEE 754 conformance.
-fsimple=1 Allows the optimizer to assume:
The IEEE 754 default rounding/trapping
modes do not change after process initialization.
Computations producing no visible result other
than potential floating-point exceptions may
be deleted. Computations with Infinity or NaNs
as operands need not propagate NaNs to their
results. For example, x*0 may be replaced by 0.
Computations do not depend on sign of zero.
-fsimple=2 Permits more aggressive floating point
optimizations that may cause programs to
produce different numeric results due to
changes in rounding. Even with -fsimple=2,
the optimizer still is not permitted to
introduce a floating point exception in a
program that otherwise produces none.
-fsingle Evaluate float expressions as single precision.
(C)
-ftrap=common Sets the IEEE 754 trapping mode to common exceptions
(C, C++, Fortran) (invalid, division by zero, and overflow).
-ftrap=%none Turns off all IEEE 754 trapping modes.
(C, C++, Fortran)
-ll2amm Include a library containing chip specific
(linker) memory routines.
-noex Do not allow C++ exceptions. A throw specification
(C++) on a function is accepted but ignored; the compiler
does not generate exception code.
-O A synomym for -xO3.
(Fortran)
-Qoption <phase> <flags>
Pass flags along to compiler phase:
f90comp Fortran first pass
iropt Global optimizer
cg Code Genetator
-Qoption cg <flags> See -Wc,<flags> below. (The code generator
(code generator) phase is addressed via -Qoption cg in
Fortran and C++; and via -Wc in C.)
-Qoption cg -Qeps:enabled=1
(code generator) See -Wc,-Qeps:enabled=1
-Qoption cg -Qeps:ws=<n>
(code generator) See -Wc,-Qeps:ws=<n>
-Qoption cg -Qgsched-T<n>
(code generator) See -Wc,-Qgsched-T<n>
-Qoption cg -Qgsched-trace_late=1
(code generator) See -Wc,-Qgsched-trace_late=1
-Qoption iropt <flags> See -W2,<flags> below. (The optimizer can
(optimizer) be addressed either via Qoption iropt in
Fortran and C++; or via -W2 in C.)
-Qoption iropt -Addint:sf=<n>
(optimizer) When considering whether to interchange loops, set memory
store operation weight to n. A higher value of n indicates
a greater performance cost for stores.
-Qoption iropt -Ainline[:cp=<n>][:cs=<n>][:inc=<n>][:irs=<n>][:mi][:recursion=1]
(optimizer) See -W2,[:cp=<n>][:cs=<n>][:inc=<n>][:irs=<n>][:mi][:recursion=1]
-Qoption iropt -Apf:llist=<n>:noinnerllist
(optimizer) Do speculative prefetching for link-list data structures:
llist=<n> perform prefetching n iterations ahead
noinnerllist do not attempt for innermost loops.
-Qoption iropt -Atile:skewp[:b<n>]
(optimizer) Perform loop tiling which is enabled by loop skewing.
Loop skewing is a transformation that transforms a
non-fully interchangeable loop nest to a fully
interchangeable loop nest. The optional b<n> sets the
tiling block size to n.
-Qoption iropt -Aujam:inner=g
(optimizer) Increase the probability that small-trip-count inner
loops will be fully unrolled.
RM_SOURCES = lapak.f90 This option allows building the benchmark 178.galgel
(SPEC tools) without its copy of the lapak sources; instead,
the lapak entry points in the sunperf library are used.
rm -rf ./feedback.profile ./SunWS_cache
(Unix) Remove any profile feedback information from previous runs.
-stackvar Force all local variables to be allocated on the stack.
(Fortran) Allocates all the local variables and arrays in
routines onto the memory stack unless otherwise
specified. This option makes these variables automatic
rather than static and provides more freedom to the
optimizer when parallelizing loops with calls to
subprograms.
Variables and arrays are local, unless they are:
o Arguments in a SUBROUTINE or FUNCTION statement
(already on stack)
o Global items in a COMMON or SAVE, or STATIC
statement
o Initialized items in a type statement or a DATA
statement, such as:
REAL X/8.0/ or DATA X/8.0/
Putting large arrays onto the stack with -stackvar can
overflow the stack causing segmentation faults.
Increasing the stack size might be required.
The initial thread executing the program has a main
stack, while each helper thread of a multithreaded
program has its own thread stack.
The default size for the main stack is about 8
Megabytes. The default helper thread stack size is 4
Megabytes on SPARC V8 platforms and 8 Megabytes on
SPARC V9 platforms.
The limit command (with no parameters) shows the
current main stack size.
Use the limit shell command to set the size (in
Kilobytes) of the main thread stack. For example, to
set the main stack size to 64 Megabytes, use a
% limit stacksize 65536
command.
You can set the stack size to be used by each slave
thread by giving the STACKSIZE environment variable a
value (in Kilobytes):
% setenv STACKSIZE 8192
sets the stack size for each slave thread to 8 Mb.
The STACKSIZE environment variable also accepts
numerical values with a suffix of either B, K, M, or G
for bytes, kilobytes, megabytes, or gigabytes
respectively. The default is kilobytes.
See the Fortran Programming Guide chapter on
parallelization for details.
See also -xcheck=stkovf to enable runtime checking for
stack overflow situations.
submit=echo 'pbind -b...' > dobmk; sh dobmk
(SPEC tools, Unix) When running multiple copies of benchmarks, the SPEC
config file feature submit is sometimes used to
cause individual jobs to be bound to specific processors:
submit= causes the SPEC tools to use this line when
submitting jobs.
echo ...> dobmk causes the generated commands to be written to a
file, namely dobmk.
pbind -b causes this copy's processes to be bound to the CPU
specified by the expression that follows it. See the
config file used in the submission for the exact
syntax, which tends to be cumbersome because of the
need to carefully quote parts of the expression.
When all expressions are evaluated, each CPU ends up
with exactly one copy of each benchmark. The pbind
expression may include:
$SPECUSERNUM: the SPEC tools-assigned number for
this copy of the benchmark.
psrinfo: find out what processors are available
grep off-line: search the psrinfo output for
information regarding off-line cpus
awk...print \$1: Pick out the line corresponding to
this copy of the benchmark and use
the CPU number mentioned at the
start of this line.
sh dobmk actually runs the benchmark.
-W<phase>,<flags> Pass flags along to compiler phase (2=optimizer,
c=code genetator).
-W2,-Abcopy Increase the probability that the compiler will
(optimizer) perform memcpy/memset transformations.
-W2,-Ainline[:cp=<n>][:cs=<n>][:inc=<n>][:irs=<n>][:mi][:recursion=1]
(optimizer) Control the optimizer's loop inliner:
(without a value) Perform Inter-Procedural Analysis (IPA) -based inlining.
cp=<n> The minimum call site frequency counter in order to
consider a routine for inlining.
cs=<n> Set inline callee size limit to n. The unit roughly
corresponds to the number of instructions.
inc=<n> The inliner is allowed to increase the size of the
program by up to n%.
irs=<n> Allow routines to increase by up to n. The unit
roughly corresponds to the number of instructions.
mi Perform maximum inlining (without considering code
size increase).
recursion=1 Allow routines that are called recursively to still
be eligible for inlining.
-W2,-crit Enable optimization of critical control paths.
(optimizer)
-W2,-Apf:llist=<n>:noinnerllist
(optimizer) Do speculative prefetching for link-list data structures:
llist=<n> perform prefetching n iterations ahead
noinnerllist do not attempt for innermost loops.
-W2,-Ashort_ldst Convert multiple short memory operations into
(optimizer) single long memory operations.
-W2,-whole Do whole program optimizations.
(optimizer)
-Wc,-Qdepgraph-early_cross_call=1
(code generator) There are several scheduling passes in the compiler.
This option allows early passes to move instructions
across call instructions.
-Wc,-Qeps:enabled=1 Use enhanced pipeline scheduling(EPS)
(code generator) and selective scheduling algorithms for
instruction scheduling.
-Wc,-Qeps:ws=<n> Set the EPS window size, that is, the number
(code generator) of instructions it will consider across all
paths when trying to find independent
instructions to schedule a parallel group.
Larger values may result in better run time,
at the cost of increased compile time.
-Wc,-Qgsched-T<n> Sets the aggressiveness of the trace
(code generator) formation, where n is 4, 5, or 6.
The higher the value of n, the lower
the branch probability needed to include
a basic block in a trace.
-Wc,-Qgsched-trace_late=1
(code generator) Turns on the late trace scheduler.
-Wc,-Qipa:valueprediction
(code generator) Use profile feedback data to predict values and attempt
to generate faster code along these control paths,
even at the expense of possibly slower code along paths
leading to different values. Correct code is generated
for all paths.
-Wc,-Qlp=<n>[-av=<n>][-t=<n>][-fa=<n>][-fl=<n>]
(code generator) Control irregular loop prefetching:
lp=<n> Turns the module on (1) or off (0)
(default is on for F90; off for C/C++)
-av=<n> Sets the prefetch look ahead distance, in bytes.
Default is 256.
-t=<n> Sets the number of attempts at prefetching. If not
specified, t=2 if -xprefetch_level=3 has been set;
otherwise, defaults to t=1.
-fa=<n> 1=Force user settings to override internally computed values.
-fl=<n> 1=Force the optimization to be turned on for all languages.
-Wc,-Qms_pipe-pref Turn off prefetching within modulo scheduling.
(code generator)
-xalias_level=[basic|std|strong]
(C) Allows the compiler to perform type-based alias analysis
at the specified alias level:
basic Assume that memory references that involve
different C basic types do not alias each other.
std Assume aliasing rules described in the ISO 1999 C
standard.
strong In addition to the restrictions at the std level,
assume that pointers of type char * are used only
to access an object of type char; and assume that
there are no interior pointers.
-xalias_level=compatible
(C++) Allows the compiler to assume that
layout-incompatible types are not aliased.
-xarch=<a> Limit the set of instructions the compiler may use
(C, C++, Fortran) to generic, generic64, native, native64, v7, v8a,
v8, v8plus, v8plusa, v8plusb, v9, v9a, v9b.
Typical settings include:
UltraSPARC-II, 32-bit mode: v8plusa
UltraSPARC-II, 64-bit mode: v9a
UltraSPARC-III, 32-bit mode: v8plusb
UltraSPARC-III, 64-bit mode: v9b
For more information, see the Fortran User's Guide
at docs.sun.com
-xbuiltin=%all Substitute intrinsic functions or inline system
(C, C++) functions where profitable for performance.
-xchip=<c> Specifies the target processor for use by the
(C, C++, Fortran) optimizer. c must be one of: generic, native,
old, super, super2, micro, micro2, hyper, hyper2,
powerup, ultra, ultra2, ultra2i, ultra3, ultra3cu,
ultra3i, ultra4, 386, 486, pentium, pentium_pro,
pentium3, pentium4
ultra3
Optimize for the UltraSPARC(TM) III processor.
ultra3cu
Optimize for the UltraSPARC IIIcu processor.
ultra4
Optimize for the UltraSPARC IV processor.
-xcache=<c> Defines the cache properties for use by the
(C, C++, Fortran) optimizer. c must be one of the following:
native (set parameters for the host environment)
* s1/l1/a1
* s1/l1/a1:s2/l2/a2
* s1/l1/a1:s2/l2/a2:s3/l3/a3
The si/li/ai are defined as follows:
si The size of the data cache
at level i, in kilobytes.
li The line size of the data cache
at level i, in bytes.
ai The associativety of the data cache
at level i.
-xdepend Analyze loops for inter-iteration data dependencies,
(C, Fortran) and do loop restructuring.
-xipo[=n] Perform optimizations across all object files in the
(C, C++, Fortran) link step:
0=off
1=on
2=performs whole-program detection and analysis
-xlibmil Use inline expansion for math library, libm.
(C, C++, Fortran)
-xlibmopt Select the optimized math library.
(C++, Fortran)
-xlic_lib=sunperf Link with Sun supplied licensed sunperf library.
(C, C++, Fortran)
-xlinkopt Perform link-time optimizations, such as branch
(C, C++, Fortran) optimization and cache coloring.
-xmemalign=ab This command specifes the maximum assumed
(C, C++, Fortran) memory alignment and the behavior of misaligned
data accesses.
For memory accesses where the alignment is
determinable at compile time, the compiler
generates the appropriate load/store instruction
sequence for that alignment of data.
For memory accesses where the alignment cannot
be determined at compile time, the compiler must
assume an alignment to generate the needed
load/store sequence.
Use the -xmemalign option to specify the maximum
memory alignment of data to be assumed by the
compiler in these indeterminable situations. You
can also specify the error behavior to be followed
at run-time when a misaligned memory access does
take place.
Values
Accepted values for a are:
1 Assume at most 1 byte alignment.
2 Assume at most 2 byte alignment.
4 Assume at most 4 byte alignment.
8 Assume at most 8 byte alignment.
16 Assume at most 16 byte alignment.
Accepted values for b are:
i Interpret access and continue execution.
s Raise signal SIGBUS.
f Raise signal SIGBUS for alignments less
than or equal to 4, otherwise interpret
access and continue execution.
You must also specify -xmemalign whenever you want to
link to an object file that was compiled with the value
of b set to either i or f. For a complete list of com-
piler options that must be specified at both compile
time and at link time, see the C User's Guide.
Defaults
The default is -xmemalign=8i for all v8 architectures.
The default for all v9 architectures is -xmemalign=8s.
Specifying -xmemalign is equivalent to specifying
-xmemalign=1i.
-xO<n> Specify optimization level n:
(C, C++, Fortran)
-xO1 Does only basic local optimizations (peephole).
-xO2 Do basic local and global optimizations, such as
induction variable elimination, common
subexpression elimination, constant propogation,
register allocation, and basic block merging.
-xO3 Add global optimizations at the function level,
loop unrolling, and software pipelining.
-xO4 Adds automatic inlining of functions in the
same file.
-xO5 Uses optmization algorithms that may take
significantly more compilation time or that
do not have as high a probability of improving
execution time, such as speculative code motion.
-xpad=common[:<n>] If multiple same-sized arrays are placed in common,
(Fortran) insert padding between them for better use of cache.
n specifies the amount of padding to apply, in units
that are the same size as the array elements. If no
parameter is specified then the compiler selects one
automatically.
-xpad=local Pad local variables, for better use of cache.
(Fortran)
-xpagesize=<n> Set the preferred page size for running the program.
(C, C++, Fortran)
-xprefetch=auto,explicit
(C, C++, Fortran) Allow generation of prefetch instructions. -xprefetch and
-xprefetch=yes is a synonym for -xprefetch=auto,explicit.
-xprefetch=latx:<n> Adjust the compiler's assumptions about prefetch latency
(C, C++, Fortran) by the specified factor. Typically values in the range of
0.5 to 2.0 will be useful. A lower number might indicate
that data will usually be cache resident; a higher number
might indicate a relatively larger gap between the
processor speed and the memory speed (compared to the
assumptions built into the compiler).
-xprefetch=no%auto Turn off prefetch instruction generation.
(C, C++, Fortran)
-xprefetch_level=<n> Control the level of searching that the compiler does
(C, C++, Fortran) for prefetch opportunities by setting n to 1, 2, or 3,
where higher numbers mean to do more searching.
The default is 2.
-xprofile=collect:./feedback
(C, C++, Fortran) Collect profile data for feedback-directed optimization,
and store it in a sub directory of the current directory,
named ./feedback.
-xprofile=use:./feedback
(C, C++, Fortran) Use data collected for profile feedback. Look for it in
a subdirectory of the current directory, named ./feedback.
-xrestrict Treat pointer-valued function parameters as
(C) restricted pointers.
-xtarget=[system_name] Selects options appropriate for the system where
(C, C++, Fortran) the compile is taking place, including architecture,
chip, and cache sizes. (These can also be controlled
separately, via -xarch, -xchip, and -xcache, respectively.)
-xunroll=n Specifies whether or not the compiler optimizes
(C, C++, Fortran) (unrolls) loops. n is a positive integer. When n is
1, it is a command and the compiler unrolls no loops.
When n is greater than 1, -xunroll=n merely suggests
to the compiler that it unroll loops n times.
-xvector Allow the compiler to transform math library calls within
(C, Fortran) loops into calls to the vector math library.
-----------------------------------------------------------------------------------
[2] Environment Variables
Flag Remark
-----------------------------------------------------------------------------------
LD_LIBRARY_PATH=<p> Specify the locations to resolve dynamic link dependencies.
LD_PRELOAD=mpss.so.1 Allow use of the mpss.so.1 shared object, which provides
a means by which preferred stack and/or heap page sizes
can be selected.
MPSSHEAP=<n> Specify the preferred page size for heap. The specified
page size is applied to all created processe.
MPSSSTACK=<n> Specify the preferred page size for stack. The specified
page size is applied to all created processe.
ulimit -s unlimited Allow stack size to grow without limit.
-----------------------------------------------------------------------------------
[3] Kernel Parameters (/etc/system)
System Tunable Remark
-----------------------------------------------------------------------------------
autoup The frequency of file system sync operations.
consistent_coloring Controls the page coloring policy. It can be set to
one of the following:
0 (default) dynamic (uses various vaddr bits)
1 static (virtual=paddr)
tune_t_fsflushr The number of seconds between fsflush invocations for
checking dirty memory.
memscrub_period_sec Period of execution of the memory patrol daemon in
units of seconds. This daemon periodically scans
memory to confirm no ECC error.
--------------------------------------------------------------------------------
[4] Commands for feedback control
Command Remark
--------------------------------------------------------------------------------
fdo_pre0 = rm -rf `pwd`/..feedback.profile
fdo_pre0 = rm -rf `pwd`/SunWS_cache
remove the profile data generated at the last
feedback-optimized compilation.