请选择 进入手机版 | 继续访问电脑版

因斯福论坛

搜索
查看: 14768|回复: 0

Intel C++ compile options

[复制链接]
发表于 2013-4-9 10:55:44 | 显示全部楼层 |阅读模式
                          Intel(R) C++ Compiler Help
                          ==========================
  Intel(R) Compiler includes compiler options that optimize for instruction
  sets that are available in both Intel(R) and non-Intel microprocessors, but
  may perform additional optimizations for Intel microprocessors than for
  non-Intel microprocessors.  In addition, certain compiler options for
  Intel(R) Compiler are reserved for Intel microprocessors.  For a detailed
  description of these compiler options, including the instructions they
  implicate, please refer to "Intel(R) Compiler User and Reference Guides >
  Compiler Options."
  usage: icl [options] file1 [file2 ...] [/link linker_options]
     where options represents zero or more compiler options
     fileN is a C/C++ source (.c .cc .cpp .cxx .i), assembly (.asm),
     object (.obj), static library (.lib), or other linkable file
     linker_options represents zero or more linker options
Notes
-----
1. Most Microsoft* Visual C++* compiler options are supported; a warning is
   printed for most unsupported options.  The precise behavior of performance
   options does not always match that of the Microsoft Visual C++ compiler.
2. Intel C++ compiler options may be placed in your icl.cfg file.
3. Most options beginning with /Q are specific to the Intel C++ compiler:
   (*I) indicates other options specific to the Intel C++ compiler
   (*M) indicates /Q options supported by the Microsoft Visual C++ compiler
   Some options listed are only available on a specific system
   i32    indicates the feature is available on systems based on IA-32
          architecture
   i64em  indicates the feature is available on systems using Intel(R) 64
          architecture
                             Compiler Option List
                             --------------------
Optimization
------------
/O1       optimize for maximum speed, but disable some optimizations which
          increase code size for a small speed benefit
/O2       optimize for maximum speed (DEFAULT)
/O3       optimize for maximum speed and enable more aggressive optimizations
          that may not improve performance on some programs
/Ox       enable maximum optimizations (same as /O2)
/Os       enable speed optimizations, but disable some optimizations which
          increase code size for small speed benefit (overrides /Ot)
/Ot       enable speed optimizations (overrides /Os)
/Od       disable optimizations
/Oi[-]    enable/disable inline expansion of intrinsic functions
/Oy[-]    enable/disable using EBP as a general purpose register (no frame
          pointer) (i32 only)
/fast     enable /QxHOST /O3 /Qipo /Qprec-div-
          options set by /fast cannot be overridden with the exception of
          /QxHOST, list options separately to change behavior
/Oa[-]    assume no aliasing in program
/Ow[-]    assume no aliasing within functions, but assume aliasing across calls
Code Generation
---------------
/Qx<code>
          generate specialized code to run exclusively on processors
          indicated by <code> as described below
            SSE2    May generate Intel(R) SSE2 and SSE instructions for Intel
                    processors.  Optimizes for the Intel NetBurst(R)
                    microarchitecture.
            SSE3    May generate Intel(R) SSE3, SSE2, and SSE instructions for
                    Intel processors.  Optimizes for the enhanced Pentium(R) M
                    processor microarchitecture and Intel NetBurst(R)
                    microarchitecture.
            SSSE3   May generate Intel(R) SSSE3, SSE3, SSE2, and SSE
                    instructions for Intel processors.  Optimizes for the
                    Intel(R) Core(TM) microarchitecture.
            SSE4.1  May generate Intel(R) SSE4 Vectorizing Compiler and Media
                    Accelerator instructions for Intel processors.  May
                    generate Intel(R) SSSE3, SSE3, SSE2, and SSE instructions
                    and it may optimize for Intel(R) 45nm Hi-k next generation
                    Intel Core(TM) microarchitecture.
            SSE4.2  May generate Intel(R) SSE4 Efficient Accelerated String
                    and Text Processing instructions supported by Intel(R)
                    Core(TM) i7 processors.  May generate Intel(R) SSE4
                    Vectorizing Compiler and Media Accelerator, Intel(R) SSSE3,
                    SSE3, SSE2, and SSE instructions and it may optimize for
                    the Intel(R) Core(TM) processor family.
            AVX     May generate Intel(R) Advanced Vector Extensions (Intel(R)
                    AVX), Intel(R) SSE4.2, SSE4.1, SSSE3, SSE3,
                    SSE2, and SSE instructions for Intel(R) processors.
                    Optimizes for a future Intel processor.
            CORE-AVX2
                    May generate Intel(R) Advanced Vector Extensions 2
                    (Intel(R) AVX2), Intel(R) AVX, SSE4.2, SSE4.1, SSSE3, SSE3,
                    SSE2, and SSE instructions for Intel(R) processors.
                    Optimizes for a future Intel processor.
            CORE-AVX-I
                    May generate Intel(R) Advanced Vector Extensions (Intel(R)
                    AVX), including instructions in Intel(R) Core 2(TM)
                    processors in process technology smaller than 32nm,
                    Intel(R) SSE4.2, SSE4.1, SSSE3, SSE3, SSE2, and SSE
                    instructions for Intel(R) processors. Optimizes for a
                    future Intel processor.
            SSSE3_ATOM
                    May generate MOVBE instructions for Intel processors,
                    depending on the setting of option /Qinstruction.
                    May also generate Intel(R) SSSE3, SSE3, SSE2, and SSE
                    instructions for Intel processors. Optimizes for the
                    Intel(R) Atom(TM) processor and Intel(R) Centrino(R)
                    Atom(TM) Processor Technology.
/QxHost   generate instructions for the highest instruction set and processor
          available on the compilation host machine
/Qax<code1>[,<code2>,...]
          generate code specialized for processors specified by <codes>
          while also generating generic IA-32 instructions.  
          <codes> includes one or more of the following:
            SSE2    May generate Intel(R) SSE2 and SSE instructions for Intel
                    processors.
            SSE3    May generate Intel(R) SSE3, SSE2, and SSE instructions for
                    Intel processors.
            SSSE3   May generate Intel(R) SSSE3, SSE3, SSE2, and SSE
                    instructions for Intel processors.
            SSE4.1  May generate Intel(R) SSE4.1, SSSE3, SSE3, SSE2, and SSE
                   instructions for Intel processors.
            SSE4.2  May generate Intel(R) SSE4.2, SSE4.1, SSSE3, SSE3, SSE2,
                    and SSE instructions for Intel processors.
            AVX     May generate Intel(R) Advanced Vector Extensions (Intel(R)
                    AVX), Intel(R) SSE4.2, SSE4.1, SSSE3, SSE3,
                    SSE2, and SSE instructions for Intel(R) processors.
            CORE-AVX2
                    May generate Intel(R) Advanced Vector Extensions 2
                    (Intel(R) AVX2), Intel(R) AVX, SSE4.2, SSE4.1, SSSE3, SSE3,
                    SSE2, and SSE instructions for Intel(R) processors.
            CORE-AVX-I
                    May generate Intel(R) Advanced Vector Extensions (Intel(R)
                    AVX), including instructions in Intel(R) Core 2(TM)
                    processors in process technology smaller than 32nm,
                    Intel(R) SSE4.2, SSE4.1, SSSE3, SSE3, SSE2, and SSE
                    instructions for Intel(R) processors.
/arch:<code>
          generate specialized code to optimize for processors indicated by
          <code> as described below
            SSE2    May generate Intel(R) SSE2 and SSE instructions
            SSE3    May generate Intel(R) SSE3, SSE2 and SSE instructions
            SSSE3   May generate Intel(R) SSSE3, SSE3, SSE2 and SSE
                    instructions
            SSE4.1  May generate Intel(R) SSE4.1, SSSE3, SSE3, SSE2 and SSE
                    instructions
            SSE4.2  May generate Intel(R) SSE4.2, SSE4.1, SSSE3, SSE3, SSE2 and
                    SSE instructions
            AVX     May generate Intel(R) AVX, SSE4.2, SSE4.1, SSSE3, SSE3,
                    SSE2 and SSE instructions
            IA32    generate x86/x87 generic code that is compatible with IA-32
                    architecture.  Disables any default extended instruction
                    settings and any previously set extended instruction
                    settings.
            SSE     same as /arch:IA32
/Qinstruction:<keyword>
          Refine instruction set output for the selected target processor
            [no]movbe  - Do/do not generate MOVBE instructions with SSSE3_ATOM
                          (requires /QxSSSE3_ATOM)
           
/GR[-]    enable/disable C++ RTTI
/Qcxx-features
          enable standard C++ features (/GX /GR)
/EHa      enable asynchronous C++ exception handling model
/EHs      enable synchronous C++ exception handling model
/EHc      assume extern "C" functions do not throw exceptions
/Qsafeseh[-]
          Registers exceptions for safe exception handling (DEFAULT)
/Gd       make __cdecl the default calling convention
/Gr       make __fastcall the default calling convention
/Gz       make __stdcall the default calling convention
/Qregcall
          make __regcall the default calling convention
/hotpatch[:n]
          generate padding bytes for function entries to enable image
          hotpatching. If specified, use 'n' as the padding.
/Qextend-arguments:[32|64]
          By default, unprototyped scalar integer arguments are passed
          in 32-bits (sign-extended if necessary).
          On Intel(R) 64, unprototyped scalar integer arguments may be
          extended to 64-bits.
Interprocedural Optimization (IPO)
----------------------------------
/Qip[-]   enable(DEFAULT)/disable single-file IP optimization
          within files
/Qipo[n]  enable multi-file IP optimization between files
/Qipo-c   generate a multi-file object file (ipo_out.obj)
/Qipo-S   generate a multi-file assembly file (ipo_out.asm)
/Qip-no-inlining
          disable full and partial inlining
/Qip-no-pinlining
          disable partial inlining
/Qipo-separate
          create one object file for every source file (overrides /Qipo[n])
/Qipo-jobs<n>
          specify the number of jobs to be executed simultaneously during the
          IPO link phase
Advanced Optimizations
----------------------
/Qunroll[n]
          set maximum number of times to unroll loops.  Omit n to use default
          heuristics.  Use n=0 to disable the loop unroller
/Qunroll-aggressive[-]
          enables more aggressive unrolling heuristics
/Qopt-args-in-regs:<keyword>
          Determines whether calls to routines are optimized by passing
          arguments in registers instead of on the stack.
          keywords: none, seen(DEFAULT), all(Linux only)
/Qscalar-rep[-]
          enable(DEFAULT)/disable scalar replacement (requires /O3)
/Qansi-alias[-]
          enable/disable(DEFAULT) use of ANSI aliasing rules optimizations;
          user asserts that the program adheres to these rules
/Qansi-alias-check[-]
          enable(DEFAULT)/disable ANSI alias checking when using /Qansi-alias
/Qcomplex-limited-range[-]
          enable/disable(DEFAULT) the use of the basic algebraic expansions of
          some complex arithmetic operations.  This can allow for some
          performance improvement in programs which use a lot of complex
          arithmetic at the loss of some exponent range.
/Qalias-const[-]
          enable/disable(DEFAULT) a heuristic stating that if two arguments to
          a function have pointer type, a pointer to const does not alias a
          pointer to non-const. Also known as the input/output buffer rule, it
          assumes that input and output buffer arguments do not overlap.
/Qalias-args[-]
          enable(DEFAULT)/disable C/C++ rule that function arguments may be
          aliased; when disabling the rule, the user asserts that this is safe
/Qopt-multi-version-aggressive[-]
          enables more aggressive multi-versioning to check for pointer
          aliasing and scalar replacement
/Qopt-ra-region-strategy[:<keyword>]
          select the method that the register allocator uses to partition each
          routine into regions
            routine - one region per routine
            block   - one region per block
            trace   - one region per trace
            loop    - one region per loop
            default - compiler selects best option
/Qvec[-]  enables(DEFAULT)/disables vectorization
/Qvec-guard-write[-]
          enables cache/bandwidth optimization for stores under conditionals
          within vector loops
/Qvec-threshold[n]
          sets a threshold for the vectorization of loops based on the
          probability of profitable execution of the vectorized loop in
          parallel
/Qopt-malloc-options:{0|1|2|3|4}
          specify malloc configuration parameters.  Specifying a non-zero <n>
          value will cause alternate configuration parameters to be set for
          how malloc allocates and frees memory
/Qopt-jump-tables:<arg>
          control the generation of jump tables
            default - let the compiler decide when a jump table, a series of
                      if-then-else constructs or a combination is generated
            large   - generate jump tables up to a certain pre-defined size
                      (64K entries)
            <n>     - generate jump tables up to <n> in size
          use /Qopt-jump-tables- to lower switch statements as chains of
          if-then-else constructs
/Qopt-block-factor:<n>
          specify blocking factor for loop blocking
/Qfreestanding
          compile in a freestanding environment where the standard library
          may not be present
/Qopt-streaming-stores:<arg>
          specifies whether streaming stores are generated
            always - enables generation of streaming stores under the
            assumption that the application is memory bound
            auto   - compiler decides when streaming stores are used (DEFAULT)
            never  - disables generation of streaming stores
/Qipp[:<arg>]
          link some or all of the Intel(R) Integrated Performance Primitives
          (Intel(R) IPP) libraries and bring in the associated headers
            common        - link using the main libraries set.  This is the
                            default value when /Qipp is specified
            crypto        - link using the main libraries set and the crypto
                            library
            gen           - link using the main libraries set and the fixed
                            size generated functions library
/Qipp-link:<arg>
          choose whether to link with static or dynamic libraries to support
          Intel(R) Integrated Performance Primitives (Intel(R) IPP)
            dynamic       - link using the dynamic libraries set.  This is the
                            default value when /Qipp is specified on Windows
            static        - link using the static libraries set.  This is the
                            default value when /Qipp is specified on Linux
            static_thread - link using the optional threaded libraries set
                            Note: these are provided in a separate install kit
/Qmkl[:<arg>]
          link to the Intel(R) Math Kernel Library (Intel(R) MKL) and bring
          in the associated headers
            parallel   - link using the threaded Intel(R) MKL libraries. This
                         is the default when /Qmkl is specified
            sequential - link using the non-threaded Intel(R) MKL libraries
            cluster    - link using the Intel(R) MKL Cluster libraries plus
                         the sequential Intel(R) MKL libraries
/Qtbb     link to the Intel(R) Threading Building Blocks (Intel(R) TBB)
          libraries and bring in the associated headers
/Qopt-subscript-in-range[-]
          assumes no overflows in the intermediate computation of the
          subscripts
/Quse-intel-optimized-headers[-]
          take advantage of the optimized header files
/Qcilk-serialize
          run a Cilk program as a C/C++ serialized program
/Qvla[-]  enable/disable(DEFAULT) Variable Length Arrays (C99 feature)
/Qintel-extensions[-]
          enable(DEFAULT)/disable C/C++ language extensions such as array
          notation, Intel(R) Cilk(TM) Plus language extensions, and
          support for decimal floating-point types.
/Qopt-matmul[-]
          replace matrix multiplication with calls to intrinsics and threading
          libraries for improved performance (DEFAULT at /O3 /Qparallel)
/Qsimd[-]
          enables(DEFAULT)/disables vectorization using simd pragma
/Qguide-opts:<arg>
          tells the compiler to analyze certain code and generate
          recommendations that may improve optimizations
/Qguide-file[:<filename>]
          causes the results of guide to be output to a file
/Qguide-file-append[:<filename>]
          causes the results of guide to be appended to a file
/Qguide[:<level>]
          lets you set a level (1 - 4) of guidance for auto-vectorization,
          auto-parallelization, and data transformation (DEFAULT is 4 when the
          option is specified)
/Qguide-data-trans[:<level>]
          lets you set a level (1 - 4) of guidance for data transformation
          (DEFAULT is 4 when the option is specified)
/Qguide-par[:<level>]
          lets you set a level (1 - 4) of guidance for auto-parallelization
          (DEFAULT is 4 when the option is specified)
/Qguide-vec[:<level>]
          lets you set a level (1 - 4) of guidance for auto-vectorization
          (DEFAULT is 4 when the option is specified)
/Qguide-profile:<[file|dir]>[,[file|dir],...]
          specify a loop profiler data file (or set of files in a directory)
          when using the /Qguide option
/Qopt-mem-layout-trans[:<level>]
          controls the level of memory layout transformations performed by the
          compiler
           0 - disable memory layout transformations (same as
               /Qopt-mem-layout-trans-)
           1 - enable basic memory layout transformations
           2 - enable more memory layout transformations  (DEFAULT when the
               option is specified)
           3 - enable aggressive memory layout transformations
/Qopt-prefetch[:n]
          enable levels of prefetch insertion, where 0 disables.
          n may be 0 through 4 inclusive.  Default is 2.
/Qopt-prefetch-
          disable(DEFAULT) prefetch insertion.  Equivalent to /Qopt-prefetch:0
Profile Guided Optimization (PGO)
---------------------------------
/Qprof-dir <dir>
          specify directory for profiling output files (*.dyn and *.dpi)
/Qprof-src-root <dir>
          specify project root directory for application source files to
          enable relative path resolution during profile feedback on sources
          below that directory
/Qprof-src-root-cwd
          specify the current directory as the project root directory for
          application source files to enable relative path resolution during
          profile feedback on sources below that directory
/Qprof-src-dir[-]
          specify whether directory names of sources should be
          considered when looking up profile records within the .dpi file
/Qprof-file <file>
          specify file name for profiling summary file
/Qprof-data-order[-]
          enable/disable(DEFAULT) static data ordering with profiling
/Qprof-func-order[-]
          enable/disable(DEFAULT) function ordering with profiling
/Qprof-gen[:keyword]
          instrument program for profiling.
          Optional keyword may be srcpos or globdata
/Qprof-gen-
          disable profiling instrumentation
/Qprof-use[:<arg>]
          enable use of profiling information during optimization
            weighted  - invokes profmerge with -weighted option to scale data
                        based on run durations
            [no]merge - enable(default)/disable the invocation of the profmerge
                        tool
/Qprof-use-
          disable use of profiling information during optimization
/Qcov-gen
          instrument program for profiling
/Qcov-dir <dir>
          specify directory for profiling output files (*.dyn and *.dpi)
/Qcov-file <file>
          specify file name for profiling summary file
/Qfnsplit[-]
          enable/disable function splitting (enabled with /Qprof-use)
/Qinstrument-functions[-]
          determine whether function entry and exit points are instrumented
/Qprof-hotness-threshold:<val>
          set the hotness threshold for function grouping and function ordering
          val indicates percentage of functions to be placed in hot region.
          This option requires /Qprof-use
           and /Qprof-func-order
         
/Qprof-value-profiling:<arg>[,<arg>,...]
          limit value profiling
            none      - inhibit all types of value profiling
            nodivide  - inhibit value profiling of non-compile time constants
                        used in division or remainder operations
            noindcall - inhibit value profiling of function addresses at
                        indirect call sites
/Qprofile-functions
          enable instrumentation in generated code for collecting
          function execution time profiles
/Qprofile-loops:<arg>
          enable instrumentation in generated code for collecting
          loop execution time profiles
             inner    - instrument inner loops
             outer    - instrument outer loops
             all      - instrument all loops
/Qprofile-loops-report:<arg>
          Control the level of instrumentation inserted for reporting loop
          execution profiles
             1    - report loop times
             2    - report loop times and iteration counts
Optimization Reports
--------------------
/Qvec-report[n]
          control amount of vectorizer diagnostic information
            n=0    no diagnostic information
            n=1    indicate vectorized loops (DEFAULT when enabled)
            n=2    indicate vectorized/non-vectorized loops
            n=3    indicate vectorized/non-vectorized loops and prohibiting
                   data dependence information
            n=4    indicate non-vectorized loops
            n=5    indicate non-vectorized loops and prohibiting data
                   dependence information
            n=6    indicate vectorized/non-vectorized loops with greater
                   details and prohibiting data dependence information
            n=7    indicate vector code quality message ids and data values
                   for vectorized loops
/Qopt-report[:n]
          generate an optimization report to stderr
            0   disable optimization report output
            1   minimum report output
            2   medium output (DEFAULT when enabled)
            3   maximum report output
/Qopt-report-file:<file>
          specify the filename for the generated report
/Qopt-report-phase:<phase>
          specify the phase that reports are generated against
/Qopt-report-routine:<name>
          reports on routines containing the given name
/Qopt-report-help
          display the optimization phases available for reporting
/Qtcheck[:mode]
          enable analysis of threaded applications (requires Intel(R) Thread
          Checker; cannot be used with compiler alone)
            tci - instruments a program to perform a thread-count-independent
                  analysis
            tcd - instruments a program to perform a thread-count-dependent
                  analysis (DEFAULT when mode is not used)
            api - instruments a program at the api-imports level
/Qtcollect[:<lib>]
          inserts instrumentation probes calling the Intel(R) Trace Collector
          API.  The library <lib>.lib is linked in the default being VT.lib
          (requires Intel(R) Trace Collector)
/Qtcollect-filter:file
          Enable or disable the instrumentation of specified functions.
          (requires Intel(R) Trace Collector)
OpenMP* and Parallel Processing
------------------------------
/Qopenmp  enable the compiler to generate multi-threaded code based on the
          OpenMP* directives (same as /openmp)
/Qopenmp-stubs
          enables the user to compile OpenMP programs in sequential mode.  The
          OpenMP directives are ignored and a stub OpenMP library is linked
          (sequential)
/Qopenmp-report{0|1|2}
          control the OpenMP parallelizer diagnostic level
/Qopenmp-lib:<ver>
          choose which OpenMP library version to link with
            compat - use the Microsoft compatible OpenMP run-time libraries
                     (DEFAULT)
/Qopenmp-task:<arg>
          choose which OpenMP tasking model to support
            omp   - support OpenMP 3.0 tasking (DEFAULT)
            intel - support Intel taskqueuing
/Qopenmp-threadprivate:<ver>
          choose which threadprivate implementation to use
            compat - use the Microsoft compatible thread local storage
            legacy - use the Intel compatible implementation
                     (DEFAULT)
/Qparallel
          enable the auto-parallelizer to generate multi-threaded code for
          loops that can be safely executed in parallel
/Qpar-report{0|1|2|3}
          control the auto-parallelizer diagnostic level
/Qpar-threshold[n]
          set threshold for the auto-parallelization of loops where n is an
          integer from 0 to 100
/Qpar-runtime-control[n]
          Control parallelizer to generate runtime check code for effective
          automatic parallelization.
            n=0    no runtime check based auto-parallelization
            n=1    generate runtime check code under conservative mode
                   (DEFAULT when enabled)
            n=2    generate runtime check code under heuristic mode
            n=3    generate runtime check code under aggressive mode
/Qpar-schedule-static[:n]
          Specifies a scheduling algorithm for DO loop iteration.
          Divides iterations into contiguous pieces.  Size n if
          specified, equal sized pieces if not.
/Qpar-schedule-static_balanced[:n]
          Divides iterations into even-sized chunks.  Size n if
          specified, equal sized pieces if not.
/Qpar-schedule-static-steal[:n]
          Divides iterations into even-sized chunks, but allows
          threads to steal parts of chunks from neighboring threads
/Qpar-schedule-dynamic[:n]
          Specifies a scheduling algorithm for DO loop iteration.
          Assigns iterations to threads in chunks dynamically.
          Chunk size is n iterations if specified, otherwise 1.
/Qpar-schedule-guided[:n]
          Specifies a scheduling algorithm for DO loop iteration.
          Indicates a minimum number of iterations.  If specified,
          n is the minimum number, otherwise 1.
/Qpar-schedule-guided-analytical[:n]
          Divides iterations by using exponential distribution or
          dynamic distributions.
/Qpar-schedule-runtime
          Specifies a scheduling algorithm for DO loop iteration.
          Defers the scheduling decision until runtime.
/Qpar-schedule-auto
          Lets the compiler or run-time system determine the
          scheduling algorithm.
/Qpar-adjust-stack<n>
          perform fiber-based main thread stack adjustment
/Qpar-affinity=[<modifier>,...]<type>[,<permute>][,<offset>]
          tune application performance by setting different thread affinity
/Qpar-num-threads=<n>
          tune application performance by setting different number of threads
/Qparallel-source-info[:n]
          enable(DEFAULT)/disable the emission of source location information
          for parallel code generation with OpenMP and auto-parallelization
            0 - disable (same as /Qparallel-source-info-)
            1 - emit routine name and line information (DEFAULT)
            2 - emit path, file, routine name and line information
/Qpar     same as /Qparallel
Floating Point
--------------
/fp:<name>
          enable <name> floating point model variation
            except[-] - enable/disable floating point semantics
            fast[=1|2]  - enables more aggressive floating point optimizations
            precise     - allows value-safe optimizations
            source      - enables intermediates in source precision
            strict      - enables /fp:precise /fp:except, disables
                          contractions and enables pragma stdc fenv_access
            double      - rounds intermediates in 53-bit (double) precision
            extended    - rounds intermediates in 64-bit (extended) precision
/Qfp-speculation:<mode>
          enable floating point speculations with the following <mode>
          conditions:
            fast   - speculate floating point operations (DEFAULT)
            safe   - speculate only when safe
            strict - same as off
            off    - disables speculation of floating-point operations
/Qpc32    set internal FPU precision to 24 bit significand
/Qpc64    set internal FPU precision to 53 bit significand (DEFAULT)
/Qpc80    set internal FPU precision to 64 bit significand
/Qprec    improve floating-point precision (speed impact less than /Op)
/Qprec-sqrt[-]
          determine if certain square root optimizations are enabled
/Qprec-div[-]
          improve precision of FP divides (some speed impact)
/Qfast-transcendentals[-]
          generate a faster version of the transcendental functions
/Qfp-port[-]
          round fp results at assignments and casts (some speed impact)
/Qfp-stack-check
          enable fp stack checking after every function/procedure call
/Qrcd     rounding mode to enable fast float-to-int conversions
/Qftz[-]  enable/disable flush denormal results to zero
/Qfma[-]  enable/disable the combining of floating point multiplies and
          add/subtract operations
/Qfp-trap:<arg>[,<arg>,...]
          control floating point traps at program start.  <arg> can be of the
          following values
            [no]divzero   - [Do not] trap on division by zero
            [no]inexact   - [Do not] trap on inexact result
            [no]invalid   - [Do not] trap on invalid operation
            [no]overflow  - [Do not] trap on overflow
            [no]underflow - [Do not] trap on underflow
            [no]denormal  - [Do not] trap on denormal
            all           - enable trap on all of the above
            none          - trap on none of the above
            common        - trap on most commonly used IEEE traps
                            (invalid, division by zero, overflow)
/Qfp-trap-all:<arg>[,<arg>,...]
          control floating point traps in every routine.  <arg> can be of the
          values specified in /Qfp-trap
Inlining
--------
/Ob<n>    control inline expansion:
            n=0  disable inlining
            n=1  inline functions declared with __inline, and perform C++
                 inlining
            n=2  inline any function, at the compiler's discretion
/Qinline-min-size:<n>
          set size limit for inlining small routines
/Qinline-min-size-
          no size limit for inlining small routines
/Qinline-max-size:<n>
          set size limit for inlining large routines
/Qinline-max-size-
          no size limit for inlining large routines
/Qinline-max-total-size:<n>
          maximum increase in size for inline function expansion
/Qinline-max-total-size-
          no size limit for inline function expansion
/Qinline-max-per-routine:<n>
          maximum number of inline instances in any function
/Qinline-max-per-routine-
          no maximum number of inline instances in any function
/Qinline-max-per-compile:<n>
          maximum number of inline instances in the current compilation
/Qinline-max-per-compile-
          no maximum number of inline instances in the current compilation
/Qinline-factor:<n>
          set inlining upper limits by n percentage
/Qinline-factor-
          do not set set inlining upper limits
/Qinline-forceinline
          treat inline routines as forceinline
/Qinline-dllimport
          allow(DEFAULT)/disallow functions declared __declspec(dllimport) to
          be inlined
/Qinline-calloc
          directs the compiler to inline calloc() calls as malloc()/memset()
Output, Debug, PCH
------------------
/c        compile to object (.obj) only, do not link
/S        compile to assembly (.asm) only, do not link
/FAs      produce assembly file with optional source annotations
/FAc      produce assembly file with optional code annotations
/FA       produce assembly file
/Fa[file]
          name assembly file (or directory for multiple files; i.e. /FaMYDIR\)
/Fo[file]
          name object file (or directory for multiple files; i.e. /FoMYDIR\)
/Fe[file]
          name executable file or directory
/Fp[file]
          name precompiled header file
/Zi, /ZI, /Z7
          produce symbolic debug information in object file (implies /Od when
          another optimization option is not explicitly set)
/debug[:keyword]
          enable debug information and control output of enhanced debug
          information
            keywords:  all, full, minimal, none, [no]inline-debug-info
                       [no]expr-source-pos
/Qtrapuv  trap uninitialized variables
/RTCu     report use of variable that was not initialized
/RTC1     equivalent to /RTCsu
/RTCs     enable stack frame runtime checks
/RTCc     enable checks for converting to smaller types
/Qmap-opts
          enable option mapping tool
/Yc[file]
          create precompiled header file
/Yu[file]
          use precompiled header file
/Y-       ignore all other precompiled header file options
/Qpchi[-]
          enable(DEFAULT)/disable PCH coexistence, to reduce build time
/FD       generate file dependencies
/FR[file]
          name extended source browser file
/Qcheck-pointers:<arg>
          Specifies what type of bounds checking occurs. Possible values are:
            none   - Disables bounds checking. This is the default.
            rw     - Checks bounds for reads and writes through pointers.
            write  - Checks bounds for writes through pointers only.
          This switch is only valid with Intel(R) Parallel Studio XE
/Qcheck-pointers-dangling:<arg>
          Specifies what type of dangling pointer checking occurs. Possible
          values are:
            none   - Disables dangling pointer checking. This is the default.
            heap   - Check dangling references on heap.
            stack  - Check dangling references on stack.
            all    - Check dangling references on both heap and stack.
          This switch is only valid with Intel(R) Parallel Studio XE
/Qcheck-pointers-undimensioned[-]
          Bounds checking occurs for memory access through arrays that are
          declared without dimensions. This checking occurs for both
          dimensioned and undimensioned arrays.
          This switch is only valid with Intel(R) Parallel Studio XE
Preprocessor
------------
/QA<name>[<val>]
          create an assertion <name> having value <val>
/C        do not strip comments
/D<name>[{=|#}<text>]
          define macro
/U<name>  remove predefined macro
/E        preprocess to stdout
/EP       preprocess to stdout, omitting #line directives
/EP /P    preprocess to file, omitting #line directives
/P        preprocess to file
/u        remove all predefined macros
/FI<file>
          force inclusion of file
/I<dir>   add directory to include file search path
/X        remove standard directories from include file search path
/QH       print include file order
/QM       generate makefile dependency information
/QMM      similar to /QM, but do not include system header files
/QMG      similar to /QM, but treat missing header files as generated files
/QMD      preprocess and compile, generating output file containing dependency
          information ending with extension .d
/QMMD     similar to /QMD, but do not include system header files
/QMF<file>
          generate makefile dependency information in file (must specify /QM
          or /QMM)
/QMT<target>
          change the default target rule for dependency generation
/QdM      output macro definitions in effect after preprocessing (use with /E)
/QdD      same as /QdM, but output #define directives in preprocessed source
/QdN      same as /QdD, but #define directives contain only macro names
Component Control
-----------------
/Qoption,<tool>,<opts>
          pass options <opts> to tool specified by <tool>
/Qlocation,<tool>,<dir>
          set <dir> as the location of tool specified by <tool>
Language
--------
/Qstd=<std>
          enable language support for <std>, as described below
            c99   enable C99 support for C programs
            c++11 enable C++11 experimental support for C++ programs
            c++0x same as c++11
/Za       disable Microsoft Visual C++ compiler language extensions
/Zg       enable the generation of function prototypes
/Zp[n]    specify alignment constraint for structures where
          n=1,2,4,8,16. 16 is the default
/Zc:arg1[,arg2]
          specify standard conformance for the following arguments:
            forScope[-] enforce standard compliance in for loop scope
            wchar_t[-]  specify that wchar_t is a native data type
            auto[-]  enforce compliance to the new standard meaning for auto
/Zs       perform syntax and semantic checking only (no object file produced)
/vd{0|1}  disable or enable hidden vtordisp field in C++ objects
/vm[x]    specify representation method for C++ pointers to members
/J        change default char type to unsigned
/noBool   suppresses the C++ bool, true, and false keywords
/Qtemplate-depth[n]
          control the depth in which recursive templates are expanded
/Qrestrict[-]
          enable/disable the 'restrict' keyword for disambiguating pointers
/Qhelp-pragma
          output supported pragmas and their usage syntax
/check:<keyword>[,<keyword>,...]
          check run-time conditions.
          keywords:  [no]conversions, [no]stack, [no]uninit
Compiler Diagnostics
--------------------
/w        disable all warnings
/W0       print only errors
/W1, /W2, /W3
          print only errors and warnings (DEFAULT)
/W4, /Wall
          print errors, warnings and most remarks
/W5       print all errors, warnings and remarks
/WL       print brief one-line diagnostics
/WX       force warnings to be reported as errors
/Werror-all
          force warnings and currently enabled remarks to be reported as errors
/Wcheck   enable more strict diagnostics
/Wport    issue portability diagnostics
/Qeffc++  enable effective C++ diagnostic warnings
/Qwo<L1>[,<L2>,...]
          issue diagnostics L1 through LN only once
/Wp64     print diagnostics for 64-bit porting
/Qdiag-enable:<v1>[,<v2>,...]
          enable the specified diagnostics or diagnostic groups
/Qdiag-disable:<v1>[,<v2>,...]
          disable the specified diagnostics or diagnostic groups
          where <vN> may be individual diagnostic numbers or group names.
          where group names include:
              sc[n]      - perform source code analysis: n=1 for critical
                           errors, n=2 for all errors and n=3 for all errors
                           and warnings
  
  sc-
  {full|concise|precise} - perform static analysis and determine the analysis
                           mode.
                           Full mode - attempts to find all program weaknesses,
                           even at the expense of more false positives.
                           Concise mode - attempts to reduce false positives
                           somewhat more than reducing false negatives.
                           Precise mode - attempts to avoid all false positives
                           Default: full if /Qdiag-enable:sc{[1|2|3]} is
                           present; otherwise None
                           (static analysis diagnostics are disabled).
                           
              sc-include - perform source code analysis on include files
              
          sc-single-file - This option tells static analysis to process each
                           file individually.
                           Default: OFF
                sc-enums - This option tells static analysis to treat
                           enumeration variables as known values equal to any
                           one of the associated enumeration literals.
                           Default: OFF
              sc-parallel[n] - perform analysis of parallelization in
                               source code: n=1 for critical errors,
                               n=2 for errors, n=3 for all errors and
                               warnings
              thread     - enable diagnostics to aid in thread-enabling source
              port-linux - enable diagnostics for Microsoft extensions that
                           may cause errors when porting to Linux/MacOS
              warn   - diagnostic messages that have "warning" severity level.
              error  - diagnostic messages that have "error" severity level.
              remark - diagnostic messages that are remarks or comments.
              vec    - diagnostic messages issued by the vectorizer.
              par    - diagnostic messages issued by the auto-parallelizer
              openmp - diagnostic messages issued by the OpenMP* parallelizer.
              cpu-dispatch Specifies the CPU dispatch remarks.
/Qdiag-error:<v1>[,<v2>,...]
          output the specified diagnostics or diagnostic groups as errors
/Qdiag-warning:<v1>[,<v2>,...]
          output the specified diagnostics or diagnostic groups as warnings
/Qdiag-remark:<v1>[,<v2>,...]
          output the the specified diagnostics or diagnostic groups as remarks
/Qdiag-dump
          display the currently enabled diagnostic messages to stdout or to a
          specified diagnostic output file.
/Qdiag-sc-dir:<dir>
          <dir> directory where diagnostics from static analysis are created,
          rather than current working directory.
/Qdiag-file[:<file>]
          <file> where diagnostics are emitted to.  Not specifying this causes
          messages to be output to stderr
/Qdiag-file-append[:<file>]
          <file> where diagnostics are emitted to. When <file> already exists,
          output is appended to the file
/Qdiag-id-numbers[-]
          enable(DEFAULT)/disable the diagnostic specifiers to be output in
          numeric form
/Qdiag-error-limit:<num>
          specify the maximum number of errors emitted
/Qdiag-once:<v1>[,<v2>,...]
          issue diagnostics v1 through vN only once
/Wpch-messages[-]
          enable(DEFAULT)/disable PCH information messages
Miscellaneous
-------------
/nologo   do not display compiler version information
/Qsox[:<keyword>[,keyword]]
          enable saving of compiler options, version and additional information
          in the executable.  Use /Qsox- to disable(DEFAULT)
            profile - include profiling data
            inline  - include inlining information
/Qsave-temps
          store the intermediate files in current directory and name them
          based on the source file.  Only saves files that are generated by
          default
/watch:<keyword>
          tells the driver to output processing information
            keywords: all, none (same as /nowatch), [no]source,
                      [no]cmd
/nowatch  suppress processing information output (DEFAULT)
/Tc<file>
          compile file as C source
/Tp<file>
          compile file as C++ source
/TC<file>
          compile all source or unrecognized file types as C source files
/TP<file>
          compile all source or unrecognized file types as C++ source files
/showIncludes
          emit include file names
/Qmultibyte-chars[-]
          provide support for multi-byte characters
/MP[<n>]  create multiple processes that can be used to compile large numbers
          of source files at the same time
/bigobj   generate objects with increased address capacity
Data
----
/Zp[n]    specify alignment constraint for structures (n=1,2,4,8,16
          /Zp16 DEFAULT)
/GS       enable overflow security checks.  
          /GS- disables (DEFAULT)
/GF       enable read-only string pooling optimization
/GA       optimize for Windows application (assume .exe)
/GT       enable fiber-safe thread local storage
/Qsfalign8
          may align stack for functions with 8 or 16 byte vars (DEFAULT)
/Qsfalign16
          may align stack for functions with 16 byte vars
/Qsfalign
          force stack alignment for all functions
/Qsfalign-
          disable stack alignment for all functions
/Qfnalign[-]
          align the start of functions to an optimal machine-dependent value.
          When disabled (DEFAULT) align on a 2-byte boundary
/Qfnalign:[2|16]
          align the start of functions on a 2 (DEFAULT) or 16 byte boundary
/Qglobal-hoist[-]
          enable(DEFAULT)/disable external globals are load safe
/Qkeep-static-consts[-]
          enable/disable(DEFAULT) emission of static const variables even
          when not referenced
/Qlong-double
          enable 80-bit 'long double'
/Qnobss-init
          disable placement of zero-initialized variables in BSS (use DATA)
/Qzero-initialized-in-bss[-]
          put explicitly zero initialized variables into the DATA section
          instead of the BSS section
/Qimf-absolute-error:value[:funclist]
          define the maximum allowable absolute error for math library
          function results
            value    - a positive, floating-point number conforming to the
                       format [digits][.digits][{e|E}[sign]digits]
            funclist - optional comma separated list of one or more math
                       library functions to which the attribute should be
                       applied
/Qimf-accuracy-bits:bits[:funclist]
          define the relative error, measured by the number of correct bits,
          for math library function results
            bits     - a positive, floating-point number
            funclist - optional comma separated list of one or more math
                       library functions to which the attribute should be
                       applied
/Qimf-arch-consistency:value[:funclist]
          ensures that the math library functions produce consistent results
          across different implementations of the same architecture
            value    - true or false
            funclist - optional comma separated list of one or more math
                       library functions to which the attribute should be
                       applied
/Qimf-max-error:ulps[:funclist]
          defines the maximum allowable relative error, measured in ulps, for
          math library function results
            ulps     - a positive, floating-point number conforming to the
                       format [digits][.digits][{e|E}[sign]digits]
            funclist - optional comma separated list of one or more math
                       library functions to which the attribute should be
                       applied
/Qimf-precision:value[:funclist]
          defines the accuracy (precision) for math library functions
            value    - defined as one of the following values
                       high   - equivalent to max-error = 0.6
                       medium - equivalent to max-error = 4 (DEFAULT)
                       low    - equivalent to accuracy-bits = 11 (single
                                precision); accuracy-bits = 26 (double
                                precision)
            funclist - optional comma separated list of one or more math
                       library functions to which the attribute should be
                       applied
Compatibility
-------------
/Qms      allow compilation of certain dubious sources
            n=0  disable Microsoft compatibility bugs
            n=1  enable most Microsoft compatibilities (DEFAULT)
            n=2  full Microsoft compatibility
/Qvc<arg>
          enable compatibility with a specific Microsoft* Visual Studio version
            9   - Microsoft* Visual Studio 2008 compatibility
            10  - Microsoft* Visual Studio 2010 compatibility
            11  - Microsoft* Visual Studio 2012 compatibility
/Qgcc-dialect:<version>
          enable a limited gcc-compatible dialect on Windows. The gcc
          version may be specified as 440, 450 or 460.
            Example: /Qgcc-dialect:440 -> gcc 4.4 compatibility
Linking/Linker
--------------
/link     specify that all options following '/link' are for the linker
/F<n>     set the stack reserve amount specified to the linker
/LD[d]    produce a DLL instead of an EXE ('d' = debug version)
/MD[d]    use dynamically-loaded, multithread C runtime
/MT[d]    use statically-linked, multithread C runtime (DEFAULT with
          Microsoft Visual Studio 2005 and later)
/ML[d]    use statically-linked, single thread C runtime (only valid in
          Microsoft Visual Studio 2003 environment)
/Zl       omit library names from object file
Deprecated Options
------------------
/Qinline-debug-info      use /debug:inline-debug-info
/Gf                      use /GF
/ML[d]                   upgrade to /MT[d]
/Quse-asm                No replacement
/Qprof-genx              use /Qprof-gen:srcpos
/Qdiag-enable:sv[<n>]    use /Qdiag-enable:sc[<n>]
/Qdiag-enable:sv-include use /Qdiag-enable:sc-include
/Qdiag-sv                use /Qdiag-enable:sc[<n>]
/Qdiag-sv-error          use /Qdiag-disable:warning
/Qdiag-sv-include        use /Qdiag-enable:sc-include
/Qdiag-sv-level          No replacement
/Qdiag-sv-sup            use /Qdiag-disable:<v1>[,<v2>,...]
/Qtprofile               No replacement
/arch:SSE                use /arch:IA32
/QxK                     upgrade to /arch:SSE2
/QaxK                    upgrade to /arch:SSE2
/QxW                     use /arch:SSE2
/QaxW                    use /arch:SSE2
/QxN                     use /QxSSE2
/QaxN                    use /QaxSSE2
/QxP                     use /QxSSE3
/QaxP                    use /QaxSSE3
/QxT                     use /QxSSSE3
/QaxT                    use /QaxSSSE3
/QxS                     use /QxSSE4.1
/QaxS                    use /QaxSSE4.1
/QxH                     use /QxSSE4.2
/QaxH                    use /QaxSSE4.2
/QxO                     use /arch:SSE3
/Qvc7.1                  No replacement
/QIfist                  use /Qrcd
/QxSSE3_ATOM             use /QxSSSE3_ATOM
/Qrct                    No replacement
/Op                      use /fp:precise
/Fr                      use /FR
/GX                      use /EHsc
/GZ                      use /RTC1
/H                       No replacement
/QA-                     use /u
/Qc99                    use /Qstd=c99
/Qwn                     use /Qdiag-error-limit:<n>
/Qwd                     use /Qdiag-disable
/Qwe                     use /Qdiag-error
/Qww                     use /Qdiag-warning
/Qwr                     use /Qdiag-remark
/Qwo                     use /Qdiag-once
/Yd                      use /Zi, /ZI, /Z7
/Zd                      use /debug:minimal
/Ze                      No replacement
/Zg                      No replacement
/help, /? [category]   print full or category help message
Valid categories include
       advanced        - Advanced Optimizations
       codegen         - Code Generation
       compatibility   - Compatibility
       component       - Component Control
       data            - Data
       deprecated      - Deprecated Options
       diagnostics     - Compiler Diagnostics
       float           - Floating Point
       help            - Help
       inline          - Inlining
       ipo             - Interprocedural Optimization (IPO)
       language        - Language
       link            - Linking/Linker
       misc            - Miscellaneous
       opt             - Optimization
       output          - Output
       pgo             - Profile Guided Optimization (PGO)
       preproc         - Preprocessor
       reports         - Optimization Reports
       openmp          - OpenMP and Parallel Processing
Copyright (C) 1985-2013, Intel Corporation.  All rights reserved.
* Other names and brands may be claimed as the property of others.

回复

使用道具 举报

您需要登录后才可以回帖 登录 | 立即注册

本版积分规则

小黑屋|因斯福论坛  

GMT+8, 2022-8-20 12:55

Powered by Discuz! X3.4

© 2001-2017 Comsenz Inc.

快速回复 返回顶部 返回列表