GNU Compiler Collection Flags for SPEC CPU

Do not rely on language constraints to derive bounds for the number of iterations of a loop.

Allow the compiler to perform optimizations that may introduce new data races on stores, without proving that the variable cannot be concurrently accessed by other threads. Does not affect optimization of local data. It is safe to use this option if it is known that global data will not be accessed by multiple threads.

Examples of optimizations enabled by -fallow-store-data-races include hoisting or if-conversions that may cause a value that was already in memory to be re-written with that same value. Such re-writing is safe in a single threaded context but may be unsafe in a multi-threaded context. On some processors, enabling if-conversions is required for vectorization.

Prior to GCC 10, use --param allow-store-data-races=1 to enable the option or 0 to disable. The default is 0.

As of GCC 10, use -fallow-store-data-races to enable or -fno-allow-store-data-races to disable. The default is -fno-allow-store-data-races unless -Ofast is used, which implies -fallow-store-data-races.

Enables a range of optimizations that provide faster, though sometimes less precise, mathematical operations.

Assume that a loop with an exit will eventually take the exit and not loop indefinitely. This allows the compiler to remove loops that otherwise have no side-effects, not considering eventual endless looping as such.

ffinite-math-only, which is implied by -fast-math and -Ofast, allows optimizations for floating-point arithmetic that assume that arguments and results are not NaNs or +-Infs. Setting -fno-finite-math-only does the opposite: the compiler must prepare for the possible presence of NaNs and infinities.

Tells GCC to use the GNU semantics for "inline" functions, that is, the behavior prior to the C99 standard. This switch may resolve duplicate symbol errors, as noted in the 502.gcc_r benchmark description.

-finline-functions-called-once, which is implied by -O1, considers all "static" functions called once for inlining into their caller even if they are not marked "inline". If a call to a given function is integrated, then the function is not output as assembler code in its own right.

-fno-inline-functions-called-once inhibits this optimization.

Perform loop interchange. This flag can improve cache performance on loop nests and allow further loop optimizations to take place, such as vectorization.

-floop-interchange is enabled by -O3.

-floop-unroll-and-jam applies unroll and jam transformations on feasible loops. In a loop nest this unrolls the outer loop by some factor and fuses the resulting multiple inner loops. This flag is enabled by default at -O3. It is also enabled by -fprofile-use and -fauto-profile..

-fno-loop-unroll-and-jam disables this optimization.

Enable Link Time Optimization When invoked with source code, it generates GIMPLE (one of GCC's internal representations) and writes it to special ELF sections in the object file. When the object files are linked together, all the function bodies are read from these ELF sections and instantiated as if they had been part of the same translation unit.

Specify the partitioning algorithm used by the link-time optimizer. The value is either 1to1 to specify a partitioning mirroring the original source files or balanced to specify partitioning into equally sized chunks (whenever possible) or max to create new partition for every symbol where possible. Specifying none as an algorithm disables partitioning and streaming completely. The default value is balanced. While 1to1 can be used as an workaround for various code ordering issues, the max partitioning is intended for internal testing only. The value one specifies that exactly one partition should be used while the value none bypasses partitioning and executes the link-time optimization step directly from the WPA phase.

Do not set "errno" after calling math functions that are executed with a single instruction, e.g., "sqrt".

Omit the frame pointer in functions that don't need one. This avoids the instructions to save, set up and restore the frame pointer; on many targets it also makes an extra register available.

fomit-frame-pointer is the default at -O1 and higher.

Enable handling of OpenMP directives and generate parallel code.

Adds the plugin named $1. For example, -fplugin=dragonegg.so loads the DragonEgg plugin, which provides an LLVM backend for GCC.

For more information, see the plugings wiki page and the plugins chapter of the GCC internals manual.

Enables prefetching of arrays used in loops.

Instruments code to collect information for profile-driven feedback. Information is collected regarding both code paths and data values.

With -fprofile-use all portions of programs not executed during train run are optimized agressively for size rather than speed. In some cases it is not practical to train all possible hot paths in the program. (For example, program may contain functions specific for a given hardware and trianing may not cover all hardware configurations program is run on.) With -fprofile-partialtraining profile feedback will be ignored for all functions not executed during the train run leading them to be optimized as if they were compiled without profile feedback. This leads to better performance when train run is not representative but also leads to significantly bigger code.

Applies information from a profile run in order to improve optimization. Several optimizations are improved when profile data is available, including branch probabilities, loop peeling, and loop unrolling.

Use the specified algorithm for basic block reordering. The algorithm argument can be 'simple', which does not increase code size (except sometimes due to secondary effects like alignment), or 'stc', the "software trace cache" algorithm, which tries to put all often executed code together, minimizing the number of branches executed by making extra copies of code.

Disable optimizations for floating-point arithmetic that ignore the signedness of zero.

Enabled: Put all local arrays, even those of unknown size onto stack memory.
The -fno- form disables the behavior.

The language standards set aliasing requirements: programmers are expected to follow conventions so that the compiler can keep track of memory. If a program violates the requirements (for example, using pointer arithmetic), programs may crash, or (worse) wrong answers may be silently produced.

Unfortunately, the aliasing requirements from the standards are not always well understood.

Sometimes, the aliasing requirements are understood and nevertheless intentionally violated by smart programmers who know what they are doing, such as the programmer responsible for the inner workings of Perl storage allocation and variable handling.

The -fno-strict-aliasing switch instructs the optimizer that it must not assume that the aliasing requirements from the standard are met by the current program. You will probably need it for 500.perlbench_r and 600.perlbench_s. Note that this is an optimization switch, not a portability switch. When running SPECint2017_rate_base or SPECint2017_speed_base, you must use the same optimization switches for all the C modules in base; see https://www.spec.org/cpu2017/Docs/runrules.html#BaseFlags and https://www.spec.org/cpu2017/Docs/runrules.html#MustValidate.

There are a group of GCC optimizations invoked via -ftree-vectorize and related flags, as described at https://gcc.gnu.org/projects/tree-ssa/vectorization.html. During testing of SPEC CPU 2017, for some versions of GCC on some chips, some benchmarks did not get correct answers when the vectorizor was enabled. These problems were difficult to isolate, and it is possible that later versions of the compiler might not encounter them.

You can turn off loop vectorization with -fno-tree-loop-vectorize. Note that this is an optimization switch, not a portability switch. If it is needed, then in base you must use it consistently. See: https://www.spec.org/cpu2017/Docs/runrules.html#BaseFlags and https://www.spec.org/cpu2017/Docs/runrules.html#MustValidate.

Attempts to decompose loops in order to run them on multiple processors.

Allow certain common subexpressions to be uncombined, which may sometimes benefit other optimizations.

Tells the optimizer to unroll all loops.

Tells the optimizer to unroll loops whose number of iterations can be determined at compile time or upon entry to the loop.

The switch -funsafe-math-optimizations allows the compiler to make certain(*) aggressive assumptions, such as disregarding the programmer's intended order of operations. The run rules allow such re-ordering https://www.spec.org/cpu2017/Docs/runrules.html#reordering. The rules also point out that you must get answers that pass SPEC's validation requirements. In some cases, that will mean that some optimizations must be turned off.

-fno-unsafe-math-optimizations turns off these(*) optimizations. You may need to use this flag in order to get certain benchmarks to validate. Note that this is an optimization switch, not a portability switch. If it is needed, then in base you will need to use it consistently. See: https://www.spec.org/cpu2017/Docs/runrules.html#BaseFlags and https://www.spec.org/cpu2017/Docs/runrules.html#MustValidate.

(*) Much more detail about which optimizations is available.

Assume that the current compilation unit represents the whole program being compiled. All public functions and variables with the exception of main and those merged by attribute externally_visible become static functions and in effect are optimized more aggressively by interprocedural optimizers.

Produce debugging information.

Add the specified path to the list of paths that the linker will search for archive libraries and control scripts.

Link with libjemalloc, a fast, arena-based memory allocator.

Compiles for a 32-bit (LP32) data model.

Compiles for a 64-bit (LP64) data model.

Generate code for ilp32 (int, long, pointer 32-bit) or lp64 (int 32-bit, longs and pointers 64-bit). With ilp32, int, long int and pointer are 32-bit; with lp64, int is 32-bit, but long int and pointer are 64-bit.

On x86 systems, allows use of instructions that require the listed architecture.

On Arm systems, specifies the name of the target architecture and, optionally, one or more feature modifiers. This option has the form -march=arch{+[no]feature}

Generate code for processors that include the AVX extensions.

On aarch64 systems, mcpu sets the what kind of instructions can be used (as if by -march) and how to tune for performance (as if by -mtune).
On x86 systems, mcpu is a deprecated synonym for mtune.
On SPARC systems, mcpu sets the available instruction set.

Generate code to take advantage of fused multiply-add

       -mrecip
           This option enables use of "RCPSS" and "RSQRTSS" instructions (and
           their vectorized variants "RCPPS" and "RSQRTPS") with an additional
           Newton-Raphson step to increase precision instead of "DIVSS" and
           "SQRTSS" (and their vectorized variants) for single-precision
           floating-point arguments.  These instructions are generated only when
           -funsafe-math-optimizations is enabled together with
           -finite-math-only and -fno-trapping-math.

       -mrecip=opt
           This option controls which reciprocal estimate instructions may be
           used.  opt is a comma-separated list of options, which may be
           preceded by a ! to invert the option:

           all
               Enable all estimate instructions.

           default
               Enable the default instructions, equivalent to -mrecip.

           none
               Disable all estimate instructions, equivalent to -mno-recip.

           div Enable the approximation for scalar division.

           vec-div
               Enable the approximation for vectorized division.

           sqrt
               Enable the approximation for scalar square root.

           vec-sqrt
               Enable the approximation for vectorized square root.

           So, for example, -mrecip=all,!sqrt enables all of the reciprocal
           approximations, except for square root.
   

Allows use of instructions that require the SIMD units of the indicated type.

Tunes code based on the timing characteristics of the listed processor.

Generate code to take advantage of version 3 of the SPARC Visual Instruction Set extensions

Disregard strict standards compliance. -Ofast enables all -O3 optimizations. It also enables optimizations that are not valid for all standard-compliant programs. It turns on -ffast-math, -fallow-store-data-races (as of GCC 10), and the Fortran-specific -fstack-arrays unless -fmax-stack-var-size is specified, and -fno-protect-parens.

Many more details are available.

Increases optimization levels: the higher the number, the more optimization is done. Higher levels of optimization may require additional compilation time, in the hopes of reducing execution time. At -O, basic optimizations are performed, such as constant merging and elimination of dead code. At -O2, additional optimizations are added, such as common subexpression elimination and strict aliasing. At -O3, even more optimizations are performed, such as function inlining and vectorization.
Many more details are available.

Same as -O1

Specify growth that the early inliner can make. In effect it increases the amount of inlining for code having a large abstraction penalty.

Specifies maximal overall growth of the compilation unit caused by inlining. For example, parameter value 20 limits unit growth to 1.2 times the original size. Cold functions (either marked cold via an attribute or by profile feedback) are not accounted into the unit size.

Specifies maximal overall growth of the compilation unit caused by interprocedural constant propagation. For example, parameter value 10 limits unit growth to 1.1 times the original size.

Maximum depth of recursive cloning for self-recursive function.

IPA-CP calculates its own score of cloning profitability heuristics and performs those cloning opportunities with scores that exceed ipa-cp-eval-threshold.

The maximum number of instructions biased by probabilities of their execution that a loop may have to be unrolled. If a loop is unrolled, this parameter also determines how many times the loop code is unrolled.

When you use -finline-functions (included in -O3), a lot of functions that would otherwise not be considered for inlining by the compiler are investigated. To those functions, a different (more restrictive) limit compared to functions declared inline can be applied.

The maximum number of instructions that a loop may have to be unrolled. If a loop is unrolled, this parameter also determines how many times the loop code is unrolled.

-fPIE causes generation of position-indepedent code suitable for use in a shared library.

-fno-PIE, which is the default, disables the generation of position-independent code.

-pie produces a dynamically linked position independent executable. SPEC CPU Makefiles are not set up to support use of this option.

-no-pie, which is the default, does not generate dynamically linked position independent executables.

On systems that support dynamic linking, this overrides -pie and prevents linking with the shared libraries. On other systems, this option has no effect.

Link the C++ library statically.

Sets the language dialect to include syntax from the C99 standard, such as bool and other features used in CPU 2017 benchmarks.

Sets the language dialect to include syntax from the 1998 ISO C++ standard plus the 2003 technical corrigendum.

Sets the language dialect to include syntax from the Fortran 2003 standard.

Pretend the symbol is undefined, to force linking of library modules to define it. You can use -u multiple times with different symbols to force loading of additional library modules. E.g., "-u malloc -ljemalloc" can gurantee to link the malloc defined in jemalloc library. Without using it, the linking options order change may cause failing to link jemalloc.

Remove unused functions from the generated executable. Without this flag, on Mac OS X, you are likely to encounter duplicate symbols when linking 502.gcc_r or 602.gcc_s.

Note that this is an optimization switch, not a portability switch. If it is needed, then in base you must use it consistently. See: https://www.spec.org/cpu2017/Docs/runrules.html#BaseFlags and https://www.spec.org/cpu2017/Docs/runrules.html#MustValidate.

Add the specified directory to the runtime library search path used when linking an ELF executable with shared objects.

Add the linker flag that requests a large stack. This flag is likely to be important only to one or two of the floating point speed benchmarks. In accordance with the rules for Base, it is set for all of fpspeed in base. See: https://www.spec.org/cpu2017/Docs/runrules.html#BaseFlags.

Set the requested page size for the program to one of the available sizes for your system - for example 2M, 4M, 1G.

Allows links to proceed even if there are multiple definitions of some symbols. This switch may resolve duplicate symbol errors, as noted in the 502.gcc_r benchmark description.

Ensure that there are no surprises if the benchmarks are run in an environment where file system metadata uses 64 bits.

Use big-endian representation for unformatted files. This is important when reading 521.wrf_r, 621.wrf_s, and 628.pop2_s data files that were originally generated in big-endian format.

Let the type "char" be signed, like "signed char".

Do not transform names of entities specified in the Fortran source file by appending underscores to them.

Let the type "char" be unsigned, like "unsigned char".

Note: this particular portability flag is included for 526.blender_r per the recommendation in its documentation - see https://www.spec.org/cpu2017/Docs/benchmarks/526.blender_r.html.

Invokes the GNU C compiler.

Invokes the GNU Fortran compiler.

Invokes the GNU C++ compiler.

This option causes all intrinsic procedures (including the GNU-specific extensions) to be accepted, such as the function imag in 521.wrf_r.

Warn, rather than generating a fatal error, when calls to external procedures have mismatches vs. the procedure definition. See also https://gcc.gnu.org/gcc-10/porting_to.html.

Place uninitialized global variables in a common block. This allows the linker to resolve all tentative definitions of the same variable in different compilation units to the same object. See also https://gcc.gnu.org/gcc-10/porting_to.html.

Allows source code in traditional (fixed-column) Fortran layout.

Downgrade some diagnostics about nonconformant code from errors to warnings, which may allow some non-conforming code to compile. Note that SPEC would be very unlikely to grant use of this option as a PORTABILITY flag; instead, SPEC would expect it to be applied to all benchmarks of a given language in base. Please see https://www.spec.org/cpu2017/Docs/runrules.html#portability and https://www.spec.org/cpu2017/Docs/runrules.html#BaseFlags.

Use pipes rather than temporary files for communication between the various stages of compilation.

Inhibit all warning messages.

Enables warnings.

Write a linker map to the named file.

Do not warn about functions defined with a return type that defaults to "int" or which return something other than what they were declared to.