编译器使用指南

1.4.1 host/device分离部分

目前编译器host编译器仅支持gcc。不支持的host端的选项如下：

• -nohdinitlist, –no-host-device-initializer-list

1.4.2 与NVCC平台相关特殊选项

本部分选项基本与NVCC工具链相关，目前编译器不支持。受影响的选项有：

• -target-dir, –target-directory

• -no-compress, –no-compress

• –no-align-double

• -nodlink, –no-device-link

• -preserve-relocs, –preserve-relocs

• -dump-callgraph, –dump-callgraph

1.4.3 宏兼容说明

NVCC中使用了许多预定义宏，nvcc wrapper中对大部分宏进行了封装支持，只有部分与机器平台强相关的宏暂不支持，下表列出了暂时不支持的宏：

• CUDACC_EWP

• NVCC_DIAG_PRAGMA_SUPPORT

1.4.4 混合编译说明

用户如使用NVCC命令行，默认情况下（-arch=sm_80）编译产物为全系列支持的混合fatbin产物; 用户可以通过特定扩展option来指定具体的编译产物：

• -arch=sm_80a 将只编译真武810E系列fatbin。

• 例外情况，-cubin不支持编译混合cubin产物，默认情况下（-arch=sm_80）将只编译真武810E系列产物。

4.2.1 文件和路径规范的选项

-cxx-isystem

• Add directory to the C++ SYSTEM include search path

-D =

• Define to (or 1 if omitted)

-fbuiltin-module-map

• Load the clang builtins module map file.

-o

• Write output to .

-fdebug-macro

• Emit macro debug information

-fdiagnostics-absolute-paths

• Print absolute paths in diagnostics

-fexceptions

• Enable support for exception handling

–fhggc-libdevice-path=

• Path to libdevice library files

-fprebuilt-module-path=

• Specify the prebuilt module path

-F

• Add directory to framework include search path

--hggc-libdevice-path=

• Specify the directory that contains the libdevice library files.

-I

• Restrict all prior -I flags to double-quoted inclusion and remove current directory from include path

-ibuiltininc

• Enable builtin #include directories even when -nostdinc is used before or after -ibuiltininc. Using -nobuiltininc after the option disables it

-idirafter

• Add directory to AFTER include search path

-iframeworkwithsysroot

• Add directory to SYSTEM framework search path, absolute paths are relative to -isysroot

-iframework

• Add directory to SYSTEM framework search path

-include-pch

• Include precompiled header file

-include

• Include file before parsing

-isystem

• Add directory to SYSTEM include search path

-isysroot

• Set the system root directory (usually /)

-I

• Add directory to the end of the list of include search paths

-l

• add the library in linking stage.

-L

• Add directory to library search path

-module-dependency-dir

• Directory to dump module dependencies to

-MF

• Write depfile output from -MMD, -MD, -MM, or -M to

-MP

• Create phony target for each dependency (other than main file)

-nobuiltininc

• Disable builtin #include directories

-nostdinc++

• Disable standard #include directories for the C++ standard library

-output-path=

• Path for output target in dependency file.

-print-resource-dir

• Print the resource directory pathname

-print-runtime-dir

• Print the directory pathname containing clangs runtime libraries

-print-search-dirs

• Print the paths used for finding libraries and programs

-stdlib++-isystem

• Use directory as the C++ standard library include path

-U

• Undefine macro

-working-directory

• Resolve file paths relative to the specified directory

-cxx-isystem

• Add directory to the C++ SYSTEM include search path

–hgas-path=

• Path to hgas (used for assemble hggc device code)

--archiver-binary=

• Specify the archiver tool path

4.2.2 指定编译阶段的选项

-c

• Only run preprocess, compile, and assemble steps

--hggc-device-only -E

• only support device preprocess

-emit-ast

• Emit Clang AST files for source inputs

-emit-interface-stubs

• Generate Interface Stub Files.

-emit-llvm

• Use the LLVM representation for assembler and object files

-emit-hggcbin

• Emit hggc bin files

-emit-merged-ifs

• Generate Interface Stub Files, emit merged text not binary.

-fas-with-hgas

• Assemble hggc device code with hgas

-dD

• Print macro definitions in -E mode in addition to normal output

-dI

• Print include directives in -E mode in addition to normal output

-dM

• Print macro definitions in -E mode instead of normal output

-D =

• Define to (or 1 if omitted)

-fc+±abi=

• C++ ABI to use. This will override the target C++ ABI.

-mstackrealign

• Force realign the stack at entry to every function.

-mix-E

• support device compile and host preprocess. The option is not supported with -fgpu-rdc.

-fchar8_t

• Enable C++ builtin type char8_t

-fcolor-diagnostics

• Enable colors in diagnostics

-fcxx-exceptions

• Enable C++ exceptions

-fdebug-macro

• Emit macro debug information

-fexceptions

• Enable support for exception handling

-fgnu89-inline

• Use the gnu89 inline semantics

-fkeep-static-consts

• Keep static const variables if unused

-fno-c+±static-destructors

• Disable C++ static destructor registration

-fno-char8_t

• Disable C++ builtin type char8_t

-fno-color-diagnostics

• Disable colors in diagnostics

-fno-debug-macro

• Do not emit macro debug information

-M

• Write a depfile containing user and system headers. Like -MD, but also implies -E and writes to stdout by default.

-MM

• Like -MMD, but also implies -E and writes to stdout by default.

-MMD

• Write a depfile containing user headers

-MD

• Write a depfile containing user and system headers.

-pedantic

• Warn on language extensions

4.2.3 指定编译器/链接器行为的选项

-x

• Treat subsequent input files as having type

-std=

• Determine the language standard. This option is currently only supported when compiling C or C++.The compiler can accept several base standards, such as c90 or c98, and GNU dialects of those standards, such as gnu90 or gnu98. When a base standard is specified, the compiler accepts all programs following that standard plus those using GNU extensions that do not contradict it. For example, -std=c90 turns off certain features of GCC that are incompatible with ISO C90, such as the “asm” and “typeof” keywords, but not other GNU extensions that do not have a meaning in ISO C90, such as omitting the middle term of a “?:” expression. On the other hand, when a GNU dialect of a standard is specified, all features supported by the compiler are enabled, even when those features change the meaning of the base standard. As a result, some strict-conforming programs may be rejected. The particular standard is used by -Wpedantic to identify which features are GNU extensions given that version of the standard. For example -std=gnu90 -Wpedantic warns about C++ style // comments, while -std=gnu99 -Wpedantic does not.option: {c11, c1x, gnu90, gnu89, gnu99, gnu9x, gnu11, gnu1x, c11, c0x, c14, c1y, c1z，c20}

• c11 c1x

ISO C11, the 2011 revision of the ISO C standard. This standard is substantially completely supported, modulo bugs, floating-point issues (mainly but not entirely relating to optional C11 features from Annexes F and G) and the optional Annexes K (Bounds-checking interfaces) and L (Analyzability). The name c1x is deprecated.

• c++11 c++0x

The 2011 ISO C++ standard plus amendments. The name c++0x is deprecated.

• c++14 c++1y

The 2014 ISO C++ standard plus amendments. The name c++1y is deprecated.

• c++1z

The next revision of the ISO C++ standard, tentatively planned for 2017. Support is highly experimental, and will almost certainly change in incompatible ways in future releases.

• c++20

The next revision of the ISO C++ standard, tentatively planned for 2020. Support is highly experimental, and will almost certainly change in incompatible ways in future releases.

-Xarch_host -g

• Generate source-level debug information for host code.

-Xarch_device -g

• Generate source-level debug information for Device code.

-gline-tables-only

• Emit debug line number tables only

-O

• Specify optimization level for host code.

-ftemplate-backtrace-limit

• Set the maximum number of entries to print in a template instantiation backtrace (0 = no limit).

-fno-exceptions

• Disable exception handling

--no-host-device-initializer-list

• Do not treat std::initializer_list membere functionn as host device.

--expt-relaxed-constexpr

• Allow host code to invoke device constexpr functions, and device code to invoke host constexpr functions.

--expt-extended-lambda

• Allow host, device annotations in lambda declarations.

-fdlto/-fno-dlto

• Perform or Disable link-time optimization of device code. Link-time optimization must be specified at both compile and link time; at compile time it stores high-level intermediate code, then at link time it links together and optimizes the intermediate code.

--host-linker-script=<use-lcs / gen-lcs>

• Generate a host linker script.

--augment-host-linker-script

• Generate an host linker script that augments an existing host linker script, need to used in combination with --host-linker-script.

--host-relocatable-link

• Generate an host linker script that can be used in host relocatable link, need to used in combination with --host-linker-script.

-ccc-gcc-name=

• specify the name for native GCC compiler.

4.2.4 传给特定编译阶段的选项

-Xassembler

• Pass to the assembler

-Xclang

• Pass to the clang compiler

-Xhggclink

• Pass to the hggclinker

-Xhggcllc

• Pass to the llc

-Xlinker

• Pass to the linker

-Xarchive

• Pass to the archiver.

-z

• Pass -z to the linker

**--options-file ,... **

• Include command line options from specified file.

-optf ,...

• Include command line options from specified file.

4.2.5 指导编译器驱动行为选项

–compatible-mode

• Use sdk compatible mode for hggc. false by default.

–default-stream-per-thread

• Normal HGGC stream per thread, does not implicitly synchronize with other streams

-fgpu-rdc

• Generate relocatable device code, also known as separate compilation mode.

-fno-gpu-rdc

• Disable relocatable device code compilation.

–hggc-compile-host-device

• Compile HGGC code for both host and device (default). Has no effect on non-HGGC compilations.

–hggc-default-device

• Make device default space specifier.

–hggc-device-only

• Compile HGGC code for device only

–hggc-enable-host-splitter

• Compile HGGC host code by using native host compiler. false by default.

–hggc-host-only

• Compile HGGC code for host only. Has no effect on non-HGGC compilations.

–hggc-link

• Link clang-offload-bundler bundles for HGGC

-Xarch_host

• set option for host code only.

-Xarch_device

• set option for device code only.

-###

• Print (but do not run) the commands to run for this compilation

-v

• Show commands to run and use verbose output

-MT

• Specify name of main file output in depfile

-save-temps=

• Save intermediate compilation results.

-save-temps

• Save intermediate compilation results.

--save-temps-path=

• Path to save intermediate compilation results.

--backend-option-file=

• Path to option files

4.2.6 通用选项

--entries=

• Specify the global entry functions for which code must be generated. May be specified more than once.

-ftime-report

• Print the time taken by each Pass.

--help

• Print help information.

-w

• disable all warning messages.

-Wreorder

• Warn when member initializers are reordered.

-Wdefault-stream-launch

• Warn when kernel launch default stream argument is used.

-Wmissing-launch-bounds

• Warn when missing launch bounds.

-Werror=

• treat specified kinds warings as errors.

• -Werror=all-warnings: treat all warnings as errors.

• -Werror=cross-execution-space-call: error when cross execution space call.

• -Werror=reorder: error when member initializers are reordered.

• -Werror=missing-launch-bounds: error when missing launch bounds

• -Werror=deprecated-declarations: error when use a deprecated entity.

• -Werror=default-stream-launch: error when kernel launch default stream argument is used.

--expt-extended-lambda

• Allow host, device annotations in lambda declarations.

4.2.7 数学库相关选项

–use_fast_math

• Make use of fast math library.–use_fast_math implies --ftz=true --prec-div=false --prec-sqrt=false --fmad=true.

–ftz {true | false}

• Control single-precision denormals support.–ftz=true flushes denormal values to zero and --ftz=false preserves denormal values. –use_fast_math implies --ftz=true.Allowed Values {true | false}DefaultThis option is set to false and ppu-clang preserves denormal values.

–prec-div {true | false}

• This option controls single-precision floating-point division and reciprocals. –prec-div=true enables the IEEE round-to-nearest mode and --prec-div=false enables the fast approximation mode. –use_fast_math implies --prec-div=false.Allowed Values {true | false}DefaultThis option is set to true and ppu-clang enables the IEEE round-to-nearest mode.

–prec-sqrt {true | false}

• This option controls single-precision floating-point square root. –prec-sqrt=true enables the IEEE round-to-nearest mode and --prec-sqrt=false enables the fast approximation mode. –use_fast_math implies --prec-sqrt=false.Allowed Values {true | false}DefaultThis option is set to true and ppu-clang enables the IEEE round-to-nearest mode.

–fmad {true | false}

• This option enables (disables) the contraction of floating-point multiplies and adds/subtracts into floating-point multiply-add operations (FMAD, FFMA, or DFMA). –use_fast_math implies --fmad=true.Allowed Values {true | false}DefaultThis option is set to true and ppu-clang enables the contraction of floating-point multiplies and adds/subtracts into floating-point multiply-add operations (FMAD, FFMA, or DFMA).

4.2.8 编译后端的option

-ppu-max-vreg-count=

• Set the maximum register number that device function can use.

-ppu-restrict

• Assert all kernel ptr is restrict pointer.

--resource-usage

• Print resource usage for kernel function.

-ppu-dlcm

• Set the default load cache policy.

-ppu-dscm

• Set the default store cache policy.

-ppu-flcm

• Force set the load cache policy.

-ppu-fscm

• Force set the store cache policy.

-ppu-warn-double-usage

• Warning if double(s) are used in an instruction.

-ppu-warn-lmem-usage

• Warning if local memory is used.

-ppu-warn-spills

• Warning if registers are spilled to local memory.

-ppu-suppress-stack-size-warning

• Suppress the warning that stack size cannot be determined.

-ppu-maxntid

• Specify the maximum number of threads that a thread block can have.

-ppu-minblockscu

• Specify the minimum number of CTAs to be mapped to an SM.

4.2.8 混合编译option

--ppu-arch==

• ppu arch，可以通过传递多个--ppu-arch option提供全系列支持的混合编译产物

• --ppu-arch==ppu001：只编译真武810E系列编译产物；

• --ppu-archppu001 --ppu-archppu0015：编译全系列的混合编译产物。