SPEC CPU2006 Flag Description for the Intel(R) C++ Compiler 11.0 for Intel(R) Fortran Compiler 11.0 for IPF Linux64 X

Intel-ic11.0-ipf SPEC CPU2006 Flag Description for the Intel(R) C++ Compiler 11.0 for Intel(R) Fortran Compiler 11.0 for IPF Linux64 X

]]>

Platform settings

One or more of the following settings may have been set. If so, the "General Notes" section of the report will say so; and you can read below to find out more about what these settings mean.

limit stacksize unlimited

Sets the stack size to n kbytes, or unlimited to allow the stack size to grow without limit.

dplace [-c cpu_numbers] [-r [l|b|t] [-v 1|2]

Dplace is used to bind a related set of processes to specific cpus or nodes to prevent process migration. In some cases, this will improve performance since a higher percentage of memory accesses will to the local node.

-c cpu numbers. Specified as a list of cpus or cpu ranges. Cpu ranges may optionally be strided: Example: "-c 1", "-c 2-4", "-c 1,4-8,3", "-c 2-8:3". The specification "-c 2-4" is equivalent to "-c 2,3,4" and "-c 2-8:3" is equivalent to 2,5,8. Ranges may also be specified in reverse order: "-c 12-8" is equivalent to 12,11,10,9,8. Cpu numbers are NOT physical cpu numbers. They are logical cpu number that are relative to the cpus that are in the set of allowed cpus as specified by the current cpuset. A cpu value of "x" (or "*"), in the argument list for -c option, indicates that binding should not be done for that process. "x" should be used only if the -e option is also used. Cpu numbers start at 0. If this option is not specified, all cpus of the current cpuset are available. The command itself (which is exec'd by dplace) is the first process to be placed by the -c cpu_numbers.
-r l|b|t. Specifies that text should be replicated on the node or nodes where the application is running. In some cases, replication will improve performance by reducing the need to make offnode memory references for code. The replication option applies to all programs placed by the dplace command. See dplace(5) for additional information on text replication. The replication options are a string of one or more of the following characters:
- l: replicate library text
- b: replicate binary (a.out) text
- t: thread round-robin option
-v 1|2. Provides the ability to run in version 1 or version 2 compatibility mode if the kernel support is available. If not specified, version 2 compatibility is selected.

Version 1 of numatools required kernel support for PAGG process placement groups. This support is no longer available in all kernel variants.

Version 2 of numatools uses a preload library to intercept calls to fork(), exec() (all variants), pthread_create() and pthread_exit(). The intercept code performs placement as part of the library call. In most cases, version 1 and version 2 are compatible. In some cases, however, a user will notice differences:
- preload libraries do not work with statically linked binaries
- preload libraries do not intercept fork() or exec() calls that come from glibc itself. Specifically, the system() call is not intercepted and no placement of tasks that result from a syst em() call will be done. In most cases, this is not an issue although you may need to adjust the if you use this option to skip tasks created by system().
In some cases, version 2 of numatools will give better performance than version 1. Assuming first-touch placement policy, in version 1 all thread-private data and a few stack pages will be located on the parent node, not the node that the task is placed on. In version 2, this memory is usually allocated local to the task's node.

dev/cpuset/memory_spread_page, /dev/cpuset/memory_spread_cache

There are two Boolean flag files per cpuset that control where the kernel allocates pages for the file system buffers and related in kernel data structures. They are called memory_spread_page and memory_spread_slab.

If the per-cpuset Boolean flag file memory_spread_page is set, the kernel will spread the file system buffers (page cache) evenly over all the nodes that the faulting task is allowed to use, instead of preferring to put those pages on the node where the task is running.

If the per-cpuset Boolean flag file memory_spread_slab is set, the kernel will spread some file system related slab caches, such as for inodes and directory entries, evenly over all the nodes that the faulting task is allowed to use; instead of preferring to put those pages on the node where the task is running.

The setting of these flags does not affect anonymous data segment or stack segment pages of a task.

When new cpusets are created, they inherit the memory spread settings of their parent.

Setting memory spreading causes allocations for the affected page or slab caches to ignore the tasks NUMA mempolicy and be spread instead. Tasks using mbind() or set_mempolicy() calls to set NUMA mempolicies will not notice any change in these calls as a result of their containing tasks memory spread settings. If memory spreading is turned off, the currently specified NUMA mempolicy once again applies to memory page allocations.

Both memory_spread_page and memory_spread_slab are Boolean flag files. By default they contain "1", meaning that the feature is on for that cpuset. If a "0" is written to that file, that turns the named feature off.

SGI ProPack for Linux

SGI ProPack is a suite of performance optimization libraries and tools for SGI Linux systems. It includes application accelerators such as NUMAtools and Flexible File I/O, parallel programming tools such as the Message Passing Toolkit, real-time performance via SGI REACT, and performance monitoring tools such as Performance Co-Pilot.

The dplace utility from the ProPack NUMAtools package is used to pin processes in CPU2006 rate runs.

]]> Invoke the Intel C++ compiler for IPF Linux64 to compile C applications

]]> Invoke the Intel C++ compiler for IPF Linux64 to compiler C++ applications

]]> Invoke the Intel Fortran compiler for IPF Linux64 Invoke the Intel C++ compiler in C99 mode for IPF Linux64 Disable all warnings. Display error messages only. For mixed-language benchmarks, tell the compiler that the main program is not written in Fortran

]]> Enables optimizations for speed and disables some optimizations that
increase code size and affect speed.
To limit code size, this option:
- Enables global optimization; this includes data-flow analysis, code motion, strength reduction and test replacement, split-lifetime analysis, and instruction scheduling.
- Disables intrinsic recognition and intrinsics inlining.
The O1 option may improve performance for applications with very large code size, many branches, and execution time not dominated by code within loops.
On IPF Linux64 platforms, -O1 disable software pipelining and global code scheduling. On Intel Itanium processors, this option also enables optimizations for server applications
(straight-line and branch-like code with a flat profile).

-unroll0, -fbuiltin, -mno-ieee-fp, -fomit-frame-pointer (same as -fp), -ffunction-sections

]]> Enables optimizations for speed. This is the generally recommended optimization level. This option also enables:
- Inlining of intrinsics
- Intra-file interprocedural optimizations, which include:
- inlining
- constant propagation
- forward substitution
- routine attribute propagation
- variable address-taken analysis
- dead static function elimination
- removal of unreferenced variables
- The following capabilities for performance gain:
- constant propagation
- copy propagation
- dead-code elimination
- global register allocation
- global instruction scheduling and control speculation
- loop unrolling
- optimized code selection
- partial redundancy elimination
- strength reduction/induction variable simplification
- variable renaming
- exception handling optimizations
- tail recursions
- peephole optimizations
- structure assignment lowering and optimizations
- dead store elimination

]]> Enables O2 optimizations plus more aggressive optimizations, such as prefetching, scalar replacement, and loop and memory access transformations. Enables optimizations for maximum speed, such as:
- Loop unrolling, including instruction scheduling
- Code replication to eliminate branches
- Padding the size of certain power-of-two arrays to allow more efficient cache use.
On Intel Itanium processors, the O3 option enables optimizations for technical computing applications (loop-intensive code):
loop optimizations and data prefetch. The O3 optimizations may not cause higher performance unless loop and memory access transformations take place. The optimizations may slow down code in some cases compared to O2 optimizations.
The O3 option is recommended for applications that have loops that heavily use floating-point calculations and process large data sets.

]]> This option enables additional interprocedural optimizations for single file compilation. These optimizations are a subset of full intra-file interprocedural optimizations. One of these optimizations enables the compiler to perform inline function expansion for calls to functions defined within the current source file. Multi-file ip optimizations that includes:
- inline function expansion
- interprocedural constant propogation
- dead code elimination
- propagation of function characteristics
- passing arguments in registers
- loop-invariant code motion

]]> The -fast option enhances execution speed across the entire program by including the following options that can improve run-time performance:

-O3 (maximum speed and high-level optimizations)

-ipo (enables interprocedural optimizations across files)

-static (link libraries statically)

To override one of the options set by -fast, specify that option after the -fast option on the command line. The options set by -fast may change from release to release.

]]> Instrument program for profiling for the first phase of two-phase profile guided otimization. This instrumentation gathers information about a program's execution paths and data values but does not gather information from hardware performance counters. The profile instrumentation also gathers data for optimizations which are unique to profile-feedback optimization.

]]> Instructs the compiler to produce a profile-optimized executable and merges available dynamic information (.dyn) files into a pgopti.dpi file. If you perform multiple executions of the instrumented program, -Qprof_use merges the dynamic information files again and overwrites the previous pgopti.dpi file.
Without any other options, the current directory is searched for .dyn files

]]> Tells the compiler the maximum number of times (n) to unroll loops. Enables inline expansion of all intrinsic functions. Disables conformance to the ANSI C and IEEE 754 standards for floating-point arithmetic.

]]> Allows use of EBP as a general-purpose register in optimizations. Places each function in its own COMDAT section.

]]> Specifies the level of inline function expansion.

Ob0 - Disables inlining of user-defined functions. Note that statement functions are always inlined.

Ob1 - Enables inlining when an inline keyword or an inline attribute is specified. Also enables inlining according to the C++ language.

Ob2 - Enables inlining of any function at the compiler's discretion.

]]> -static prevents linking with shared libraries.

]]> This option enables read only string-pooling optimization. This option enables read/write string-pooling optimization. This option disables stack-checking for routines with 4096 bytes of local variables and compiler temporaries.

]]> Enable SmartHeap library usage by forcing the linker to ignore multiple definitions Enable SmartHeap library usage by forcing the linker to ignore multiple definitions Disables the insertion of software prefetching by the compiler. Default is -prefetch. This option turns on versioning of modulo operations for certain types of operands (e.g. x%y where y is dynamically determined to be a power of 2). The default is modulo versioning off. This option may improve performance. Versioning of modulo operations commonly results in possibly large speedups for x%y where y is a power of 2. However, the optimization could hurt performance slightly if y is not a power of 2. This option tells the compiler to use more aggressive unrolling for certain loops. The default is -no-unroll-aggressive (the compiler uses less aggressive default heuristics when unrolling loops). This option may improve performance. On the Itanium architecture, this option enables additional complete unrolling for loops that have multiple exits or outer loops that have a small constant trip count. This option controls the prefetches that are issued for a memory access in the next iteration, typically done in a pointer-chasing loop. This option should improve performance. The default is -no-opt-prefetch-next-iteration (next iteration prefetch off). This option controls the prefetches that are issued before the loop is entered. These prefetches target the initial iterations of the loop. The default is -opt-prefetch-initial-values (prefetch for initial iterations on) at -O1 and higher optimization levels. This option controls the loadpair optimization. The loadpair optimization is enabled by default when -O3 is used for Itanium. -no-opt-loadpair turns the loadpair optimization off. Specifies the percentage multiplier that should be applied to all inlining options that define upper limits. The value is a positive integer specifying the percentage value. The default value is 100 (a factor of 1). Enables use of faster but slightly less accurate code sequences for math functions, including sqrt, reciprocal sqrt, divide and reciprocal. When compared to strict IEEE* precision, this option slightly reduces the accuracy of floating-point calculations performed by these functions, usually limited to the least significant digit. This option also performs reassociation transformations, which can alter the order of operations, over a larger scope. The increased reasssociation enables generation of more optimal sequences of floating-point multiply-add instructions than not using this option. Note that use of floating-point multiply-add can cause programs to produce different numerical results due to changes in rounding. Tells the compiler to assume the program does adhere to to the Fortran 95 Standard type aliasability rules (default). Tells the compiler to assume the program does adhere to the rules defined in the ISO C Standard. The default is to not assume such adherence. If your C/C++ program adheres to these rules, then -ansi-alias will allow the compiler to optimize more aggressively. If it doesn't adhere to these rules, then assuming so can cause the compiler to generate incorrect code. Do not assume arguments may be aliased. (DEFAULT = -alias-args). Tells the compiler not to assume aliasing in the program (DEFAULT = -falias). Instructs the compiler to analyze and transform the program so that 64-bit pointers are shrunk to 32-bit pointers, and 64-bit longs (on Linux) are shrunk into 32-bit longs wherever it is legal and safe to do so. In order for this option to be effective the compiler must be able to optimize using the -ipo option and must be able to analyze all library or external calls the program makes. This option requires that the size of the program executable never exceeds 2 (to the 32nd power) bytes and all data values can be represented within 32 bits. If the program can run correctly in a 32-bit system, these requirements are implicitly satisfied. If the program violates these size restrictions, unpredictable behavior might occur. The -Wl option directs the compiler to pass a list of arguments to the linker. In this case, "-z muldefs" is passed to the linker. For the Gnu linker (ld), the "-z keyword" option accepts several recognized keywords. Keyword "muldefs" allows multiple definitions. The muldefs keyword will enable, for example, linking with third party libraries like SmartHeap from Microquill. Rule to eat the paths from SmartHeap library inclusion. MicroQuill SmartHeap Library available from http://www.microquill.com/ MicroQuill SmartHeap Library available from http://www.microquill.com/ MPI library.