Clang 14.0.6 Release Notes¶
Written by the LLVM Team
Introduction¶
This document contains the release notes for the Clang C/C++/Objective-C/OpenCL frontend, part of the LLVM Compiler Infrastructure, release 14.0.6. Here we describe the status of Clang in some detail, including major improvements from the previous release and new feature work. For the general LLVM release notes, see the LLVM documentation. All LLVM releases may be downloaded from the LLVM releases web site.
For more information about Clang or LLVM, including information about the latest release, please see the Clang Web Site or the LLVM Web Site.
Note that if you are reading this file from a Git checkout or the main Clang web page, this document applies to the next release, not the current one. To see the release notes for a specific release, please see the releases page.
What’s New in Clang 14.0.6?¶
Some of the major new features and improvements to Clang are listed here. Generic improvements to Clang as a whole or to its underlying infrastructure are described first, followed by language-specific sections with improvements to Clang’s support for those languages.
Major New Features¶
Added SPIR-V triple and binary generation using external
llvm-spirvtool. For more details refer to the SPIR-V support section.Completed support of OpenCL C 3.0 and C++ for OpenCL 2021 at experimental state.
Prebuilt AIX7.2 TL5 SP3+ binary available with following notes and limitations:
C++ driver modes use the system libc++ headers. These headers are included in the optional
libc++.adt.includefileset on AIX.LTO, although not disabled, is not meaningfully functional for practical use.
Shared libraries builds (
-shared) must use explicit symbol export options and/or export lists (e.g., with-bE:) on the link step. Clang currently will not automatically generate symbol export lists as implicit linker inputs.
float.hnow exposes (in hosted mode) extensions made available from the AIX system header.
Improvements to Clang’s diagnostics¶
-Wbitwise-instead-of-logical (part of -Wbool-operation) warns about use of bitwise operators with boolean operands which have side effects.
Added diagnostic groups to control diagnostics for attribute extensions by adding groups
-Wc++N-attribute-extensions(whereNis the standard release being diagnosed against). These new groups are automatically implied when passing-Wc++N-extensions. Resolves PR33518.Support
-Wdeclaration-after-statementwith C99 and later standards, and not just C89, matching GCC’s behaviour. A notable usecase is supporting style guides that forbid mixing declarations and code, but want to move to newer C standards.
Non-comprehensive list of changes in this release¶
Maximum _ExtInt size was decreased from 16,777,215 bits to 8,388,608 bits. Motivation for this was discussed in PR51829.
Configuration file syntax extended with
<CFGDIR>token. This expands to the base path of the current config file. See Configuration files for details.The
-E -Ppreprocessor output now always omits blank lines, matching gcc behaviour. Previously, up to 8 consecutive blank lines could appear in the output.AIX platform-related predefined macros added:
_ARCH_PPC64,__HOS_AIX__,__PPC,__THW_BIG_ENDIAN__,__THW_PPC__, and__powerpc
New Compiler Flags¶
Clang plugin arguments can now be passed through the compiler driver via
-fplugin-arg-pluginname-arg, similar to GCC’s-fplugin-arg.The
-mno-bti-at-return-twiceflag will make sure a BTI instruction won’t be added after a setjmp or possible other return-twice construct (ARM backend only).The
--start-no-unused-argumentsand--end-no-unused-argumentsflags allow silencing warnings about unused arguments for only a subset of the command line arguments, keeping potential warnings for other arguments outside of such a region.-falign-loops=N(N is a power of 2) is now supported for non-LTO cases. (D106701)The
-fminimize-whitespaceflag allows removing redundant whitespace from preprocessor output (-Eflag). When combined with-P, this includes newlines. Otherwise, only indention is removed (other horizontal whitespace is always collapsed). The motivation is to improve compiler cache hit rate by becoming invariant to whitespace changes, such as reformatting using clang-format. Patches for ccache and sccache are under review.Clang now accepts “allowlist” spelling for
-objcmt-allowlist-dir-path.
Deprecated Compiler Flags¶
-Wweak-template-vtables has been deprecated and no longer has any effect. The flag will be removed in the next release.
Modified Compiler Flags¶
Support has been added for the following processors (
-mcpuidentifiers in parentheses):RISC-V SiFive E20 (
sifive-e20).RISC-V SiFive E21 (
sifive-e21).RISC-V SiFive E24 (
sifive-e24).RISC-V SiFive E34 (
sifive-e34).RISC-V SiFive S21 (
sifive-s21).RISC-V SiFive S51 (
sifive-s51).RISC-V SiFive S54 (
sifive-s54).RISC-V SiFive S76 (
sifive-s76).Arm Cortex-X1C (
cortex-x1c)Arm Cortex-X2 (
cortex-x2)Arm Cortex-A510 (
cortex-a510)Arm Cortex-A710 (
cortex-a710)
Support has been added for the following architectures (
-marchidentifiers in parentheses):Armv9-A (
armv9-a).Armv9.1-A (
armv9.1-a).Armv9.2-A (
armv9.2-a).
-rnow implies-nostdlibfor many toolchains, matching GCC. (D116843)
Removed Compiler Flags¶
The legacy
-gz=zlib-gnuand-Wa,--compress-debug-sections=zlib-gnuhave been removed.-fno-experimental-new-pass-managerhas been removed.-flegacy-pass-managercan be used as a makeshift, Using the legacy pass manager for the optimization pipeline was deprecated in 13.0.0 and will be removed after 14.0.0.
New Pragmas in Clang¶
…
Attribute Changes in Clang¶
Attributes loaded as clang plugins which are sensitive to LangOpts must now override
acceptsLangOptsinstead ofdiagLangOpts. Returning false will produce a generic “attribute ignored” diagnostic, as with clang’s built-in attributes. If plugins want to provide richer diagnostics, they can do so when the attribute is handled instead, e.g. inhandleDeclAttribute. (This was changed in order to better support attributes in code completion).__has_cpp_attribute, __has_c_attribute, __has_attribute, and __has_declspec will now macro expand their argument. This causes a change in behavior for code using
__has_cpp_attribute(__clang__::attr)(and same for__has_c_attribute) where it would previously expand to0for all attributes, but will now issue an error due to the expansion of the predefined__clang__macro.Improved handling of
__attribute__((__aligned__))on AIX to match GCC.
Windows Support¶
An MSVC compatibility workaround for C++ operator names was removed. As a result, the
<query.h>Windows SDK header may not compile out of the box. Users should use a recent SDK and pass-DQUERY_H_RESTRICTION_PERMISSIVEor pass/permissiveto disable C++ operator names altogether. See PR42427 <https://llvm.org/pr42427> for more info.Add support for MSVC-compatible
/hotpatchflag in clang-cl, and equivalent -cc1 flag-fms-hotpatch. Along with the linker flag/functionpadminthis creates executable images suitable for runtime code patching. This flag is only required for x86/x64 targets; ARM/ARM64 simply needs the linker/functionpadmin.With this addition, clang-cl can be used in live code patching scenarios, along with tools such as Live++ or Recode. Microsoft Edit and Continue isn’t currently supported.
Support for on-demand initialization of TLS variables was added.
Improved code generation for ARM, by assuming less strict alignment requirements for instructions (just like other OSes do).
Fixed using the
-m32flag in x86_64 MinGW setups, by e.g. making-m32pick i686 instead of i386, if there is no i386 sysroot, but only one for i686.Fixed passing the
--no-demangleoption through to the linker for MinGW targets.
C Language Changes in Clang¶
The value of
__STDC_VERSION__has been bumped to202000Lwhen passing-std=c2xso that it can be distinguished from C17 mode. This value is expected to change again when C23 is published.Wide multi-characters literals such as
L'ab'that would previously be interpreted asL'b'are now ill-formed in all language modes. The motivation for this change is outlined in P2362.Support for
__attribute__((error("")))and__attribute__((warning("")))function attributes have been added.The maximum allowed alignment has been increased from 2^29 to 2^32.
Clang now supports the
_BitInt(N)family of bit-precise integer types from C23. This type was previously exposed as_ExtInt(N), which is now a deprecated alias for_BitInt(N)(so diagnostics will mention_BitInteven if source uses_ExtInt)._BitInt(N)and_ExtInt(N)are the same types in all respects beyond spelling and the deprecation warning._BitInt(N)is supported as an extension in older C modes and in all C++ modes. Note: the ABI for_BitInt(N)is still in the process of being stabilized, so this type should not yet be used in interfaces that require ABI stability. The maximum width supported by Clang can be obtained from theBITINT_MAXWIDTHmacro in<limits.h>. Currently, Clang supports bit widths <= 128 because backends are not yet able to cope with some math operations (like division) on wider integer types. See PR44994 for more information.When using
asm gotowith outputs whose constraint modifier is"+", we now change the numbering of the labels to occur after hidden tied inputs for better compatibility with GCC. For better portability between different compilers and versions, symbolic references rather than numbered references should be preferred. See this thread <https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103640> for more info.Implemented WG14 N2412, which adds
*_WIDTHmacros to limits.h and stdint.h to report the bit width of various integer datatypes.The
ATOMIC_VAR_INITmacro from<stdatomic.h>is now diagnosed as deprecated in C17 and later. The diagnostic can be disabled by defining the_CLANG_DISABLE_CRT_DEPRECATION_WARNINGSmacro prior to including the header.
C++ Language Changes in Clang¶
…
C++20 Feature Support¶
The
ATOMIC_VAR_INITandATOMIC_FLAG_INITmacros from the C standard library<stdatomic.h>header are now diagnosed as deprecated in C++20 and later. Note, the behavior is specific to the inclusion of<stdatomic.h>in C++ code; the STL<atomic>header also controls the behavior of these macros and is not affected by these changes. The<stdatomic.h>diagnostic can be disabled by defining the_CLANG_DISABLE_CRT_DEPRECATION_WARNINGSmacro prior to including the header.
C++2b Feature Support¶
Implemented P1938R3: if consteval.
Implemented P2360R0: Extend init-statement to allow alias-declaration.
CUDA Language Changes in Clang¶
Clang now supports CUDA versions up to 11.5.
Default GPU architecture has been changed from sm_20 to sm_35.
Objective-C Language Changes in Clang¶
OpenCL Kernel Language Changes in Clang¶
OpenCL 3.0 is completed, but remains experimental:
Added parsing support for optionality of device-side enqueue and blocks.
Added missing support for optionality of various builtin functions:
read_writeimages, pipes, collective workgroup in the default header.read_writeimages, named address space atomics in internalopencl-c.h(enabled via-finclude-default-headerfrontend flag).
C++ for OpenCL:
Added experimental support of C++ for OpenCL 2021 as per the provisional language documentation. Support of all optional features is aligned with OpenCL 3.0.
Added
__remove_address_spaceutility (documentation available in Clang Language Extensions).Fixed address space for temporaries (to be
__private).Disallowed static kernel functions.
Fixed implicit definition of
__cpp_threadsafe_static_initmacro.
Misc changes:
Added generation of SPIR-V binaries via external
llvm-spirvtool. For more details refer to the SPIR-V support section.Added new extensions for
atomic_halfandcl_ext_float_atomics.Fixed/improved support of
vload/vstore.Fixed incorrect
as_typesupport for 3-element vector types.
ABI Changes in Clang¶
The
_ExtInt(N)extension has been standardized in C23 as_BitInt(N). The mangling of this type in C++ has accordingly changed: under the Microsoft ABI it is now mangled using the_BitIntspelling, and under the Itanium ABI it is now mangled using a dedicated production. Note: the ABI for_BitInt(N)is still in the process of being stabilized, so this type should not yet be used in interfaces that require ABI stability.
OpenMP Support in Clang¶
clang-nvlink-wrappertool introduced to support linking of cubin files archived in an archive. See Clang Nvlink Wrapper.clang-linker-wrappertool introduced to support linking using a new OpenMP target offloading method. See Clang Linker Wrapper.
CUDA Support in Clang¶
…
X86 Support in Clang¶
Support for
AVX512-FP16instructions has been added.
DWARF Support in Clang¶
The default DWARF version has increased from DWARFv4 to DWARFv5. You can opt back in to the old behavior with
-gdwarf-4or-fdebug-default-version=4. Some platforms (Darwin, Android, and SCE for instance) already opt out of this version bump as is suitable for the platform
Arm and AArch64 Support in Clang¶
The -mtune flag is no longer ignored for AArch64. It is now possible to tune code generation for a particular CPU with -mtune without setting any architectural features. For example, compiling with “-mcpu=generic -mtune=cortex-a57” will not enable any Cortex-A57 specific architecture features, but will enable certain optimizations specific to Cortex-A57 CPUs and enable the use of a more accurate scheduling model.
The –aarch64-none-elf target now uses the BareMetal driver rather than the GNU driver. Programs that depend on clang invoking GCC as the linker driver should use GCC as the linker in the build system.
The
-mbranch-protectionflag will now also work for the ARM backend.The
attribute((target("branch-protection=...)))attributes will now also work for the ARM backend.When using
-mbranch-protection=btiwith AArch64, calls to setjmp will now be followed by a BTI instruction. This is done to be compatible with setjmp implementations that return with a br instead of a ret. You can disable this behaviour using the-mno-bti-at-return-twiceoption.
SPIR-V Support in Clang¶
Added triple/target
spirv32andspirv64for 32-bit and 64-bit SPIR-V respectively.Added generation of binaries via external
llvm-spirvtool. This can now be used for HIP or OpenCL.Added linking of separate object files in SPIR-V format using external
spirv-linktool.
Floating Point Support in Clang¶
The default setting of FP contraction (FMA) is now -ffp-contract=on (for languages other than CUDA/HIP) even when optimization is off. Previously, the default behavior was equivalent to -ffp-contract=off (-ffp-contract was not set). Related to this, the switch -ffp-model=precise now implies -ffp-contract=on rather than -ffp-contract=fast, and the documentation of these features has been clarified. Previously, the documentation claimed that -ffp-model=precise was the default, but this was incorrect because the precise model implied -ffp-contract=fast, whereas the (now corrected) default behavior is -ffp-contract=on. -ffp-model=precise is now exactly the default mode of the compiler.
-fstrict-float-cast-overflow no longer has target specific behavior. Clang will saturate towards the smallest and largest representable integer values. NaNs will be converted to zero.
Internal API Changes¶
A new sugar
TypeAST node represents types accessed via a C++ using declaration. Given codeusing std::error_code; error_code x;,xhas aUsingTypewhich desugars to the previousRecordType.
Build System Changes¶
Linux distros can specify
-DCLANG_DEFAULT_PIE_ON_LINUX=Onto use-fPIEand-pieby default. This matches GCC installations on many Linux distros (configured with--enable-default-pie). (D113372)
AST Matchers¶
TypeLocAST Matchers are now available. These matchers provide helpful utilities for matchingTypeLocnodes, such as thepointerTypeLocmatcher or thehasReturnTypeLocmatcher. The addition of these matchers was made possible by changes to the handling ofTypeLocnodes that allows them to enjoy the same static type checking as other AST node kinds.LambdaCaptureAST Matchers are now available. These matchers allow for the binding ofLambdaCapturenodes. TheLambdaCapturematchers added include thelambdaCapturenode matcher, thecapturesVartraversal matcher, andcapturesThisnarrowing matcher.The
hasAnyCapturematcher now only accepts an inner matcher of typeMatcher<LambdaCapture>. The matcher originally accepted an inner matcher of typeMatcher<CXXThisExpr>orMatcher<VarDecl>.The
usingTypematcher is now available and needed to refer to types that are referred to via using C++ using declarations. The associatedUsingShadowDeclcan be matched usingthroughUsingDecland the underlyingTypewithhasUnderlyingType.hasDeclarationcontinues to see through the alias and apply to the underlying type.Added the
isConstevalmatcher to matchconstevalfunction declarations as well as if consteval and if ! consteval statements.Added the
isConstinitmatcher to matchconstinitvariable declarations.
clang-format¶
Option
AllowShortEnumsOnASingleLine: falsehas been improved, it now correctly places the opening brace according toBraceWrapping.AfterEnum.Option
AlignAfterOpenBracket: BlockIndenthas been added. If set, it will always break after an open bracket, if the parameters don’t fit on a single line. Closing brackets will be placed on a new line.Option
QualifierAlignmenthas been added in order to auto-arrange the positioning of specifiers/qualifiers const volatile static inline constexpr restrict in variable and parameter declarations to be eitherRightaligned orLeftaligned orCustomusingQualifierOrder.Option
QualifierOrderhas been added to allow the order const volatile static inline constexpr restrict to be controlled relative to the type.Option
RemoveBracesLLVMhas been added to remove optional braces of control statements for the LLVM style.Option
SeparateDefinitionBlockshas been added to insert or remove empty lines between definition blocks including functions, classes, structs, enums, and namespaces.Add a
Customstyle toSpaceBeforeParens, to better configure the space before parentheses. The custom options can be set usingSpaceBeforeParensOptions.The command line argument -style=<string> has been extended so that a specific format file at location <format_file_path> can be selected. This is supported via the syntax: -style=file:<format_file_path>.
Improved C++20 Modules and Coroutines support.
Option
AfterOverloadedOperatorhas been added inSpaceBeforeParensOptionsto allow space between overloaded operator and opening parentheses.
libclang¶
…
Static Analyzer¶
…
Undefined Behavior Sanitizer (UBSan)¶
Core Analysis Improvements¶
…
New Issues Found¶
…
Python Binding Changes¶
The following methods have been added:
…
Significant Known Problems¶
Additional Information¶
A wide variety of additional information is available on the Clang web
page. The web page contains versions of the
API documentation which are up-to-date with the Git version of
the source code. You can access versions of these documents specific to
this release by going into the “clang/docs/” directory in the Clang
tree.
If you have any questions or comments about Clang, please feel free to contact us via the mailing list.