Skip to content

Latest commit

 

History

History
249 lines (192 loc) · 9.33 KB

Preprocessor-Macros.md

File metadata and controls

249 lines (192 loc) · 9.33 KB
Raw

Overview

Preprocessor macros allow a user to write portable code in the face of non-portable features by enabling platform-specific code only when it is supported.

External Resources

See Pre-defined C/C++ Compiler Macros for an incredibly thorough list of preprocessor macros for various compilers, systems, etc.

Manual Extraction

With GCC, you can get the predefined macros using gcc -dM -E - < /dev/null.

Languages

Language Macro Details
C89 __STDC__ ANSI X3.159-1989
C90 __STDC_VERSION__ ISO/IEC 9899:1990
C94 __STDC_VERSION__ = 199409L ISO/IEC 9899-1:1994
C99 __STDC_VERSION__ = 199901L ISO/IEC 9899:1999
C11 __STDC_VERSION__ = 201112L ISO/IEC 9899:2011
C20 __STDC_VERSION__ > 201710L
C++ __cplusplus
C++98 __cplusplus = 199711L ISO/IEC 14882:1998
C++11 __cplusplus = 201103L ISO/IEC 14882:2011
C++14 __cplusplus = 201402L ISO/IEC 14882:2014
C++17 __cplusplus = 201703L ISO/IEC 14882:2017
C++20 __cplusplus > 201703L

See https://en.cppreference.com/w/cpp/feature_test for fine-grain feature testing in C++.

Language Extensions for Parallelism

Language Macro Details
UPC __UPC__ UPC Identification
UPC __UPC_DYNAMIC_THREADS__ The integer constant 1 in the dynamic THREADS translation environment.
UPC __UPC_STATIC_THREADS__ The integer constant 1 in the static THREADS translation environment.
OpenMP _OPENMP OpenMP
OpenMP _OPENMP = 200505 OpenMP 2.5
OpenMP _OPENMP = 200805 OpenMP 3.0
OpenMP _OPENMP = 201107 OpenMP 3.1
OpenMP _OPENMP = 201307 OpenMP 4.0
OpenMP _OPENMP = 201511 OpenMP 4.5
OpenMP _OPENMP = 201811 OpenMP 5.0
Cilk __cilk = 200 Cilk++

Compilers

GCC

Macro Purpose
__GNUC__ Identification
__GNUC__ Version
__GNUC_MINOR__ Revision
__GNUC_PATCHLEVEL__ Patch (introduced in version 3.0)

Clang/LLVM

Macro Purpose
__llvm__ Identification of LLVM
__clang__ Identification of Clang
__clang_major__ Clang Version
__clang_minor__ Clang Revision
__clang_patchlevel__ Clang Patch

Intel

Macro Purpose
__INTEL_COMPILER Identification
__INTEL_COMPILER Version (Format: VRP, where V = Version, R = Revision, P = Patch)
__INTEL_OFFLOAD True if Intel compiler supports offload (on by default)

Cray

Macro Purpose
_CRAYC Identification
_RELEASE Version
_RELEASE_MINOR Revision

IBM

Macro Purpose
__xlc__ Identification
__xlC__ Identification
__IBMC__ Identification
__IBMCPP__ Identification
__IBMC__ Version (Format: VRP, where V = Version, R = Revision, P = Patch)
__IBMCPP__ Version (Format: VRP, where V = Version, R = Revision, P = Patch)

NVHPC

Use nvc++ -stdpar -cuda -acc -mp -target=multicore -dM -E /dev/null | grep NV to find more...

Macro Purpose
__NVCOMPILER Identification
__NVCOMPILER_MAJOR__ Version
__NVCOMPILER_MINOR__ Revision
__NVCOMPILER_PATCHLEVEL__ Patch
__NVCOMPILER_CUDA__, _NVHPC_CUDA CUDA support enabled
_NVHPC_STDPAR_GPU, _NVHPC_STDPAR_CUDA StdPar for GPU with CUDA (i.e. -stdpar=gpu)
_NVHPC_STDPAR_MULTICORE, _NVHPC_STDPAR_OMP StdPar for CPU with OpenMP (i.e. -stdpar=multicore)
__NVCOMPILER_CUDA_ARCH__ CUDA arch (e.g. 860 for sm_86)

Note that the following PGI macros are also supported, for historical reasons.

PGI

Macro Purpose
__PGI Identification
__PGIC__ Version
__PGIC_MINOR__ Revision
__PGIC_PATCHLEVEL__ Patch

Programming Models

MPI

Macro Purpose
MPI_VERSION Version
MPI_SUBVERSION Revision
MPICH Identification of MPICH
MPICH_NUMVERSION MPICH Version for numerical comparison (See mpi.h for details)
MPICH_VERSION MPICH Version as string
OPEN_MPI Identification of OpenMPI
OMPI_MAJOR_VERSION OpenMPI Version
OMPI_MINOR_VERSION OpenMPI Revision
OMPI_RELEASE_VERSION OpenMPI Patch Level

CUDA

Macro Purpose
__CUDACC__ Identification
__CUDACC_VER__ Toolkit Version - added in CUDA 7.5, removed in CUDA 8.0 (Format: XXYYZZ, where XX = Major Version Number, YY = Minor Version Number, ZZ = Build Number)
__CUDACC_VER_MAJOR__ Toolkit Major Version - added in CUDA 7.5
__CUDACC_VER_MINOR__ Toolkit Minor Version - added in CUDA 7.5
__CUDACC_VER_BUILD__ Toolkit Build Version - added in CUDA 7.5
CUDA_VERSION API Version - defined in <cuda.h> (Format: XXYY, where XX = Major Version Number, YY = Minor Version Number)

Operating Systems

Some of these were obtained from the site linked above and are assumed to be correct.

OS Macro Details
Linux __linux__
Mac OSX __APPLE__ This may not be the best way to do this since this macro may be defined in other operating systems produced by Apple.
Mac OSX __MACH__ Distinguishes Darwin/Mach kernel from iOS.
BSD __FreeBSD__, __NetBSD__, __OpenBSD__, __bsdi__, __DragonFly__ Unverified.
AIX _AIX Unverified due to lack of system access.
Blue Gene CNK __bg__ See below for details.

TODO: Test for POSIX support?

Architectures

Cray

Manually extracted from Cray compilers (i.e. verified):

Macro Meaning
__CRAYXT Defined on the Cray XT architecture.
__CRAYXE Defined on the Cray XE architecture.
__CRAYXC Defined on the Cray XC architecture.
__CRAYXT_COMPUTE_LINUX_TARGET Defined when Cray is using Linux, as opposed to Catamount.

From Cray documentation (i.e. reputable):

Macro Meaning
__CRAY Defined as 1 on UNICOS/mp systems.
_CRAYIEEE Defined as 1 if the targeted CPU type uses IEEE floating-point format.

Blue Gene

From IBM Blue Gene/P compiler documentation and IBM Blue Gene/Q compiler documentation (i.e. reliable):

Macro Meaning
__bg__ Indicates that this is a Blue Gene platform.
__blrts, __blrts__ Indicates that the target architecture is Blue Gene/L.
__bgp__ Indicates that the architecture is Blue Gene/P.
__bgq__ Indicates that the architecture is Blue Gene/Q.
__TOS_BLRTS__ Indicates that the target architecture is Blue Gene/L.
__TOS_BGP__ Indicates that the target architecture is Blue Gene/P.
__TOS_BGQ__ Indicates that the target architecture is Blue Gene/Q.
__VECTOR4DOUBLE__ Indicates the support of vector data types on Blue Gene/Q.

The __bg__, __bgp__, and __bgq__ have been verified on both Blue Gene/P and Blue Gene/Q systems and are the recommended one in this authors opinion.

Processor Features

Intel/AMD

Suffix Macro Description
SSE __SSE__ Introduced with Pentium 3.
SSE2 __SSE2__ Introduced with Willamette (Pentium 4).
SSE3 __SSE3__ Introduced with Prescott.
SSE4.1 __SSE4_1__ Introduced with Penryn.
SSE4.2 __SSE4_2__ Introduced with Penryn.
SSSE3 __SSSE3__ Introduced with Merom.
AVX __AVX__ Introduced with Sandy Bridge.
AVX2 __AVX2__ Introduced with Haswell.
FMA __FMA__ Fused-multiple-add instructions.
FMA4 __FMA4__ See https://en.wikipedia.org/wiki/FMA_instruction_set.

AVX-512

Suffix Macro Description
F __AVX512F__ Foundation
CD __AVX512CD__ Conflict Detection Instructions (CDI)
ER __AVX512ER__ Exponential and Reciprocal Instructions (ERI)
PF __AVX512PF__ Prefetch Instructions (PFI)
DQ __AVX512DQ__ Doubleword and Quadword Instructions (DQ)
BW __AVX512BW__ Byte and Word Instructions (BW)
VL __AVX512VL__ Vector Length Extensions (VL)
IFMA __AVX512IFMA__ Integer Fused Multiply Add (IFMA)
VBMI __AVX512VBMI__ Vector Byte Manipulation Instructions (VBMI)
VBMI2 __AVX512VBMI2__ Vector Byte Manipulation Instructions 2 (VBMI2)
BITALG __AVX512BITALG__ Vector Bit Algorithms
VNNI __AVX512VNNI__ Vector Neural Network Instructions
VNNIW __AVX5124VNNIW__ Vector instructions for deep learning enhanced word variable precision
FMAPS __AVX5124FMAPS__ Vector instructions for deep learning floating-point single precision
VPOPCNT __AVX512VPOPCNTDQ__ Vectorized Hamming weight
VP2INTERSECT __AVX512VP2INTERSECT__ Compute intersection between doublewords/quadwords to a pair of mask registers

The primary source for these is the Intel® Architecture Instruction Set Extensions Programming Reference.

The intrinsics for AVX-512 are nicely documented in the Intel Intrinsics Guide.

Wikipedia's articles AVX-512 is a useful unofficial source.

LLVM source code (http://llvm.org/viewvc/llvm-project/cfe/trunk/test/Preprocessor/predefined-arch-macros.c?view=diff&r1=262200&r2=262201&pathrev=262201) is the basis for two of the entries in the above table.

Cauldron14_AVX-512_Vector_ISA_Kirill_Yukhin_20140711.pdf and Intel® AVX-512 Instructions and Their Use in the Implementation of Math Functions may be of interest.