Intel® oneAPI DPC++/C++ Compiler Developer Guide and Reference

Download PDF
ID 767253
Date 7/13/2023
Public

A newer version of this document is available. Customers should click here to go to the newest version.

Document Table of Contents
Document Table of Contents x
Intel® oneAPI DPC++/C++ Compiler Developer Guide and Reference
Intel® oneAPI DPC++/C++ Compiler Developer Guide and Reference x
Intel® oneAPI DPC++/C++ Compiler Introduction Compiler Setup Compiler Reference Compilation Optimization and Programming Compatibility and Portability Notices and Disclaimers
Intel® oneAPI DPC++/C++ Compiler Introduction x
Get Help and Support Related Information
Compiler Setup x
Use the Command Line Use Eclipse Use Microsoft Visual Studio
Use the Command Line x
Specify Component Locations Invoke the Compiler Use the Command Line on Windows File Extensions Use Makefiles for Compilation Use CMake with the Compiler Use Compiler Options Specify Compiler Files Convert Projects to Use a Selected Compiler
Use Eclipse x
Add the Compiler to Eclipse Multiversion Compiler Support Use Cheat Sheets Create a Simple Eclipse Project Makefiles Use Intel Libraries with Eclipse
Use Microsoft Visual Studio x
Create a New Project Use the Intel® oneAPI DPC++/C++ Compiler Select the Compiler Version Specify a Base Platform Toolset Use Property Pages Use Intel® Libraries with Microsoft Visual Studio* Include MPI Support Dialog Box Help
Dialog Box Help x
Options: Compilers dialog box Use Intel® oneAPI DPC++/C++ Compiler dialog box Options: Intel Libraries for oneAPI dialog box
Compiler Reference x
C/C++/SYCL Calling Conventions Compiler Options Floating-Point Operations Attributes Intrinsics Libraries Macros Pragmas Error Handling
Compiler Options x
Alphabetical Option List General Rules for Compiler Options What Appears in the Compiler Option Descriptions Optimization Options Code Generation Options Interprocedural Optimization Options Advanced Optimization Options Profile Guided Optimization Options Optimization Report Options Offload Compilation, OpenMP*, and Parallel Processing Options Floating-Point Options Inlining Options Output, Debug, and Precompiled Header Options Preprocessor Options Component Control Options Language Options Data Options Compiler Diagnostic Options Compatibility Options Linking or Linker Options Miscellaneous Options Deprecated and Removed Compiler Options Display Option Information Alternate Compiler Options Portability and GCC-Compatible Warning Options
Optimization Options x
fast fbuiltin, Oi foptimize-sibling-calls GF nolib-inline O Od Ofast Os Ot Ox
Code Generation Options x
arch ax, Qax EH fasynchronous-unwind-tables fcf-protection, Qcf-protection fdata-sections, Gw fexceptions ffunction-sections, Gy fomit-frame-pointer Gd GR guard Gv m, Qm m64, Qm64 m80387 march masm mauto-arch, Qauto-arch mbranches-within-32B-boundaries, Qbranches-within-32B-boundaries mintrinsic-promote, Qintrinsic-promote momit-leaf-frame-pointer mtune, tune regcall, Qregcall x, Qx xHost, QxHost
Interprocedural Optimization Options x
flto ipo, Qipo
Advanced Optimization Options x
ffreestanding, Qfreestanding fjump-tables fvec-peel-loops, Qvec-peel-loops fvec-remainder-loops, Qvec-remainder-loops fvec-with-mask, Qvec-with-mask ipp-link, Qipp-link mno-gather, Qgather- mno-scatter, Qscatter- qactypes, Qactypes qdaal, Qdaal qipp, Qipp qmkl, Qmkl qmkl-ilp64, Qmkl-ilp64 qopt-assume-no-loop-carried-dep, Qopt-assume-no-loop-carried-dep qopt-dynamic-align, Qopt-dynamic-align qopt-for-throughput, Qopt-for-throughput qopt-mem-layout-trans, Qopt-mem-layout-trans qopt-multiple-gather-scatter-by-shuffles, Qopt-multiple-gather-scatter-by-shuffles qopt-prefetch, Qopt-prefetch qopt-streaming-stores, Qopt-streaming-stores qtbb, Qtbb unroll, Qunroll use-intel-optimized-headers, Quse-intel-optimized-headers vec, Qvec vec-threshold, Qvec-threshold
Profile Guided Optimization Options x
fprofile-ml-use
Optimization Report Options x
qopt-report, Qopt-report qopt-report-file, Qopt-report-file qopt-report-stdout, Qopt-report-stdout
Offload Compilation, OpenMP*, and Parallel Processing Options x
device-math-lib fintelfpga fiopenmp, Qiopenmp flink-huge-device-code fno-sycl-libspirv foffload-static-lib fopenmp fopenmp-declare-target-scalar-defaultmap, Qopenmp-declare-target-scalar-defaultmap fopenmp-device-lib fopenmp-target-buffers, Qopenmp-target-buffers fopenmp-targets, Qopenmp-targets fsycl fsycl-add-targets fsycl-dead-args-optimization fsycl-device-code-split fsycl-device-lib fsycl-device-obj fsycl-device-only fsycl-early-optimizations fsycl-enable-function-pointers fsycl-esimd-force-stateless-mem fsycl-explicit-simd fsycl-force-target fsycl-help fsycl-host-compiler fsycl-host-compiler-options fsycl-id-queries-fit-in-int fsycl-instrument-device-code fsycl-link fsycl-link-huge-device-code fsycl-link-targets fsycl-max-parallel-link-jobs fsycl-optimize-non-user-code fsycl-rdc fsycl-targets fsycl-unnamed-lambda fsycl-use-bitcode ftarget-compile-fast nolibsycl qopenmp, Qopenmp qopenmp-link qopenmp-simd, Qopenmp-simd qopenmp-stubs, Qopenmp-stubs reuse-exe Wno-sycl-strict Xopenmp-target Xs Xsycl-target
Floating-Point Options x
ffp-contract fimf-absolute-error, Qimf-absolute-error fimf-accuracy-bits, Qimf-accuracy-bits fimf-arch-consistency, Qimf-arch-consistency fimf-domain-exclusion, Qimf-domain-exclusion fimf-max-error, Qimf-max-error fimf-precision, Qimf-precision fimf-use-svml, Qimf-use-svml fma, Qfma fp-model, fp fp-speculation, Qfp-speculation ftz, Qftz pc, Qpc
Inlining Options x
fgnu89-inline finline finline-functions
Output, Debug, and Precompiled Header Options x
c debug (Linux*) debug (Windows*) Fa fasm-blocks Fe Fo Fp fverbose-asm g gdwarf grecord-gcc-switches gsplit-dwarf o RTC S use-msasm Y- Yc Yu Zi, Z7, ZI
Preprocessor Options x
B C D dD, QdD dM, QdM E EP FI H, QH I idirafter imacros iprefix iquote isystem iwithprefix iwithprefixbefore M, QM MD, QMD MF, QMF MG, QMG MM, QMM MMD, QMMD MQ, QMQ MT, QMT nostdinc++ P TP U undef X
Component Control Options x
Qoption
Language Options x
ansi fno-gnu-keywords fno-operator-names fno-rtti fpermissive fshort-enums fsyntax-only, Zs funsigned-char, J std, Qstd strict-ansi vd vmg x (type option) Zc Zg Zp
Data Options x
fcommon fkeep-static-consts, Qkeep-static-consts fmath-errno fpack-struct fpic fpie fstack-protector fstack-security-check fvisibility fzero-initialized-in-bss, Qzero-initialized-in-bss GA Gs GS mcmodel Qlong-double
Compiler Diagnostic Options x
w w, W Wabi Wall Wcheck-unicode-security Wcomment Wdeprecated Weffc++, Qeffc++ Werror, WX Werror-all Wextra-tokens Wformat Wformat-security Wmain Wmissing-declarations Wmissing-prototypes Wpointer-arith Wreorder Wreturn-type Wshadow Wsign-compare Wstrict-aliasing Wstrict-prototypes Wtrigraphs Wuninitialized Wunknown-pragmas Wunused-function Wunused-variable Wwrite-strings
Compatibility Options x
gcc-toolchain vmv
Linking or Linker Options x
F (Windows*) fixed fuse-ld l L LD link MD MT nodefaultlibs no-intel-lib, Qno-intel-lib nostartfiles nostdlib pie pthread shared shared-intel shared-libgcc static static-intel static-libgcc static-libstdc++ T u (Linux*) v Wa Wl Wp Xlinker Zl
Miscellaneous Options x
dryrun dumpmachine dumpversion help nologo save-temps, Qsave-temps showIncludes sox, Qsox sysroot Tc TC Tp version watch
Floating-Point Operations x
Programming Tradeoffs in Floating-Point Applications Use Floating-Point Options Tune Compilation Accuracy Dispatching of Math Routines See Also Floating-Point Optimizations Denormal Numbers Floating-Point Environment Set the FTZ and DAZ Flags Tuning Performance IEEE Floating-Point Operations
Attributes x
align align_value allow_cpu_features const cpu_dispatch, cpu_specific target
Libraries x
Create Libraries Use Intel Shared Libraries on Linux Manage Libraries Redistribute Libraries When Deploying Applications Resolve References to Shared Libraries Intel's Memory Allocator Library SIMD Data Layout Templates Intel® C++ Class Libraries Intel's C++ Asynchronous I/O Extensions for Windows IEEE 754-2008 Binary Floating-Point Conformance Library Intel's Numeric String Conversion Library
SIMD Data Layout Templates x
Function Calls and Containers User-Level Interface Examples
Function Calls and Containers x
Construct an n_container Bounds
User-Level Interface x
SDLT Primitives soa1d_container aos1d_container n_container Bounds Accessors Proxy Objects Number Representation Indexes Convenience and Correctness
aos1d_container x
access_by
n_container x
Layouts Shape make_ n_container template function extent_d template function
Shape x
n_extent_generator
Bounds x
bounds_t sdlt::bounds Template Function n_bounds_t n_bounds_generator bounds_d Template Function
Accessors x
soa1d_container::accessor and aos1d_container::accessor soa1d_container::const_accessor and aos1d_container::const_accessor Accessor Concept
Proxy Objects x
Proxy ConstProxy
Number Representation x
aligned_offset fixed_offset
Indexes x
linear_index n_index_t n_index_generator index_d template function
Convenience and Correctness x
max_val min_val
Examples x
Efficiency with Structure of Arrays Example Complex SDLT Primitive Construction Example Forward Dependency Example Use of Offsets and Methods on a SDLT Primitive Example RGB to YUV Conversion Example
Intel® C++ Class Libraries x
C++ Classes and SIMD Operations Capabilities of C++ SIMD Classes Integer Vector Classes Floating-Point Vector Classes Classes Quick Reference Programming Example Intel's valarray Implementation
Integer Vector Classes x
Terms and Syntax Rules for Operators Assignment Operator Logical Operators Addition and Subtraction Operators Multiplication Operators Shift Operators Comparison Operators Conditional Select Operators Debug Operations Unpack Operators Pack Operators Clear MMX™ State Operator Integer Functions for Intel® Streaming SIMD Extensions Conversions between Fvec and Ivec
Floating-Point Vector Classes x
Fvec Syntax and Notation Data Alignment Conversions Constructors and Initialization Arithmetic Operators Minimum and Maximum Operators Logical Operators Compare Operators Conditional Select Operators for Fvec Classes Cacheability Support Operators Debug Operations Load and Store Operators Unpack Operators Move Mask Operators
Intel's C++ Asynchronous I/O Extensions for Windows x
Intel's C++ Asynchronous I/O Library for Windows Intel's C++ Asynchronous I/O Class for Windows
Intel's C++ Asynchronous I/O Library for Windows x
aio_read aio_write Example for aio_read and aio_write Functions aio_suspend Example for aio_suspend Function aio_error aio_return Example for aio_error and aio_return Functions aio_fsync aio_cancel Example for aio_cancel Function lio_listio Example for lio_listio Function Asynchronous I/O Function Errors
Intel's C++ Asynchronous I/O Class for Windows x
Template Class async_class get_last_operation_id wait get_status get_last_error get_error_operation_id stop_queue resume_queue clear_queue Example for Using async_class Template Class
IEEE 754-2008 Binary Floating-Point Conformance Library x
Intel® IEEE 754-2008 Binary Floating-Point Conformance Library and Usage Function List Homogeneous General-Computational Operations Functions General-Computational Operation Functions Quiet-Computational Operations Functions Signaling-Computational Operations Functions Non-Computational Operations Functions
Intel's Numeric String Conversion Library x
Use Intel's Numeric String Conversion Library Function List
Macros x
ISO Standard Predefined Macros Additional Predefined Macros Use Predefined Macros to Specify Intel® Compilers
Pragmas x
Intel-Specific Pragma Reference Intel-Supported Pragma Reference
Intel-Specific Pragma Reference x
block_loop/noblock_loop distribute_point inline, noinline, forceinline ivdep loop_count nofusion novector omp target variant dispatch ompx prefetch data prefetch/noprefetch unroll/nounroll unroll_and_jam/nounroll_and_jam vector
Compilation x
Compilation Overview Supported Environment Variables Pass Options to the Linker Specify Alternate Tools and Paths Use Configuration Files Use Response Files Global Symbols and Visibility Attributes for Linux* Save Compiler Information in Your Executable Link Debug Information Ahead of Time Compilation Device Offload Compilation Considerations Use a Third-Party Compiler as a Host Compiler for SYCL Code
Optimization and Programming x
Extensions OpenMP* Support Intel® oneAPI Level Zero Vectorization High-Level Optimization Interprocedural Optimization Methods to Optimize Code Size Intel® oneAPI DPC++/C++ Compiler Math Library
OpenMP* Support x
Add OpenMP* Support Parallel Processing Model Worksharing Using OpenMP* Control Thread Allocation OpenMP* Pragmas OpenMP* Library Support OpenMP* Advanced Issues OpenMP* Implementation-Defined Behaviors OpenMP* Examples
OpenMP* Library Support x
OpenMP* Runtime Library Routines Intel® Compiler Extension Routines to OpenMP* OpenMP* Support Libraries Use the OpenMP Libraries Thread Affinity Interface OpenMP* Memory Spaces and Allocators
Intel® oneAPI Level Zero x
Intel® oneAPI Level Zero Switch Intel® oneAPI Level Zero Backend Specification Programming with the Intel® oneAPI Level Zero Backend
Vectorization x
Automatic Vectorization Explicit Vector Programming
Automatic Vectorization x
Vectorization Programming Guidelines Use Automatic Vectorization Vectorization and Loops Loop Constructs
Explicit Vector Programming x
User-Mandated or SIMD Vectorization SIMD-Enabled Functions SIMD-Enabled Function Pointers Function Annotations and the SIMD Directive for Vectorization Explicit SIMD SYCL Extension
Interprocedural Optimization x
Use Interprocedural Optimization Performance Considerations Create a Library from IPO Objects Inline Expansion of Functions
Intel® oneAPI DPC++/C++ Compiler Math Library x
Use the Intel® oneAPI DPC++/C++ Compiler Math Library Math Function List Trigonometric Functions Hyperbolic Functions Exponential Functions Special Functions Nearest Integer Functions Remainder Functions Miscellaneous Functions Complex Functions C99 Macros
Compatibility and Portability x
Standards Conformance GCC Compatibility and Interoperability Microsoft Compatibility Port from Microsoft Visual C++* to the Intel® oneAPI DPC++/C++ Compiler Port from GCC* to the Intel® oneAPI DPC++/C++ Compiler
Port from Microsoft Visual C++* to the Intel® oneAPI DPC++/C++ Compiler x
Modify Your makefile Other Considerations
Port from GCC* to the Intel® oneAPI DPC++/C++ Compiler x
Modify Your makefile Other Considerations
Intel® oneAPI DPC++/C++ Compiler Developer Guide and Reference

Programming Tradeoffs in Floating-Point Applications

In general, the programming objectives for floating-point applications fall into the following categories:

  • Accuracy: The application produces results that are close to the correct result.
  • Reproducibility and portability: The application produces consistent results across different runs, different sets of build options, different compilers, different platforms, and different architectures.
  • Performance: The application produces fast, efficient code.

Based on the goal of an application, you will need to make tradeoffs among these objectives. For example, if you are developing a 3D graphics engine, performance may be the most important factor to consider, with reproducibility and accuracy as secondary concerns.

The default behavior of the compiler is to compile for performance. Several options are available that allow you to tune your applications based on specific objectives. Broadly speaking, there are the floating-point specific options, such as the -fp-model (Linux*) or /fp (Windows*) option, and the fast-but-low-accuracy options, such as the [Q]imf-max-error option (host only). The compiler optimizes and generates code differently when you specify these different compiler options. Select appropriate compiler options by carefully balancing your programming objectives and making tradeoffs among these objectives. Some of these options may influence the choice of math routines that are invoked.

Use Floating-Point Options

The default behavior of the compiler is to use fp-model=fast. In this mode, lower-accuracy versions of the math library functions are chosen. For host code, this only affects calls that have been vectorized. For target code, the exact effect will vary depending on the target.

For GPU devices, using fp-model=fast enables lower-accuracy versions of the functions. Lower accuracy implementations are conformant with the OpenCL 3.0 specification. SVML functions with up to four ULPs (the equivalent to using -fimf-precision=medium) are used.

For FPGA devices, using fp-model=fast enables lower-accuracy versions of the functions, but there is no specific limit on the accuracy.

With fp-model=precise, the host code will use high accuracy implementations for both scalar and SVML calls. Target devices use implementations that conform to the SYCL specification, which isn't as accurate as host implementations, but is more accurate than with fp-model=fast.

Take the following code as an example: 

float t0, t1, t2;
 ...
t0=t1+t2+4.0f+0.1f;

If you specify the -fp-model precise (Linux) or /fp:precise (Windows) option in favor of accuracy, the compiler generates the following assembly code: 

movss     xmm0, _t1 
addss     xmm0, _t2 
addss     xmm0, DWORD PTR _Cnst4.0 
addss     xmm0, DWORD PTR _Cnst0.1 
movss     DWORD PTR _t0, xmm0
The assembly code follows the same semantics as the original source code.

If you specify the -fp-model fast (Linux) or /fp:fast (Windows) option in favor of performance, the compiler generates the following assembly code: 

movss     xmm0, DWORD PTR _Cnst4.1 
addss     xmm0, DWORD PTR _t1 
addss     xmm0, DWORD PTR _t2 
movss     DWORD PTR _t0, xmm0

This code maximizes performance using Intel® Streaming SIMD Extensions (Intel® SSE) instructions and pre-computing 4.0f + 0.1f. It is not as accurate as the first implementation, due to the greater intermediate rounding error. It does not provide reproducible results because it must reorder the addition to pre-compute 4.0f + 0.1f. When fast-math is enabled, the ordering of operations is decided by the compiler. Different ordering may be used depending on the context, and not all compilers will choose the same ordering.

For many other applications, the considerations may be more complicated.

Tune Compilation Accuracy

In general, the -fp-model option provides control for accuracy. However, the compiler provides command-line options for an easy way to control the accuracy of mathematical functions and utilize performance/accuracy tradeoffs offered by the Intel math libraries that are provided with the compiler. These options are helpful in the following scenarios:

  • Use high-accuracy implementations while otherwise allowing fast-math optimizations
  • Use faster-but-less-accurate implementations while otherwise disabling fast-math optimizations

You can specify accuracy, via a command line interface, for all math functions or a selected set of math functions at a level more precise than low, medium, or high.

You specify the accuracy requirements as a set of function attributes that the compiler uses for selecting an appropriate function implementation in the math libraries. For example, use the following option to specify the relative error of two ULPs for all single, double, long double, and quad precision functions: 

-fimf-max-error=2

To specify twelve bits of accuracy for a sin function, use: 

-fimf-accuracy-bits=12:sin

To specify relative error of ten ULPs for a sin function, and four ULPs for other math functions called in the source file you are compiling, use: 

-fimf-max-error=10:sin -fimf-max-error=4

On Windows systems, the compiler defines the default value for the max-error attribute depending on the /fp option setting. In /fp:fast mode the compiler sets a max-error=4.0 for the call. Otherwise, it sets a max-error=0.6.

For high-accuracy floating-point options on host code, use the --fimf-precision option. For high-accuracy floating-point options on device code, use the -f[no-]approx-func option.

On Windows use the /fp:precise option for more accurate floating-point SYCL operations.

Note that the OpenCL* standard provides guidelines for each floating-point operation, such as cos(), that specify the maximum ULP variance that a conforming device must support. For example, for cosine, a conforming device cannot have an ULP variance higher than 4. However, with the default fast floating-point operations, the ULP variance will likely be higher than what the OpenCL standard requires.

Dispatching of Math Routines

The compiler optimizes calls to routines from the libm and svml libraries into direct CPU-specific calls, when the compilation configuration specifies the target CPU where the code is tuned, and if the set of instructions available for the code compilation is not narrower than the set of instructions available in the tuning target CPU.

Note that except in the case of functions which return correctly-rounded results (<= 0.5 ulp error), you cannot rely on being able to obtain bitwise identical results from different device types. This is mainly due to differences in the implementation of math library functions which are optimized for the available instruction set on the device.

The use of floating-point options to require high accuracy implementations of the math library routines will reduce the impact of this problem, but not eliminate it. Depending on the algorithm used by the program being compiled, small errors may be compounded.

The use of less accurate implementations may amplify the differences. For example, if the cos() function is called with a four ULP error implementation, all devices will return a result that is within four ULP of the theoretically accurate result, but there is no guarantee that two different devices will return the same result within that error range.

Parent topic: Floating-Point Operations

See Also

Level Two Title