GCC 14 Release Series
Changes, New Features, and Fixes

This page is a "brief" summary of some of the huge number of improvements in GCC 14. You may also want to check out our Porting to GCC 14 page and the full GCC documentation.

Caveats

• C: Support for the GCC extension, a structure containing a C99 flexible array member, or a union containing such a structure, is not the last field of another structure, is deprecated. Refer to Zero Length Arrays. Any code relying on this extension should be modifed to ensure that C99 flexible array members only end up at the ends of structures. Please use the warning option -Wflex-array-member-not-at-end to identify all such cases in the source code and modify them.
• C: Certain warnings about are now errors, see Porting to GCC 14 for details.
• -fcf-protection=[full|branch|return|none|check] is refactored, to override -fcf-protection, -fcf-protection=none needs to be added and then with -fcf-protection=xxx.
• Support for the ia64*-*- target ports which have been unmaintained for quite a while has been declared obsolete in GCC 14. The next release of GCC will have their sources permanently removed.
• Support for the nios2*-*- target ports has also been declared obsolete in GCC 14, and the sources will also be removed in the next release of GCC.
• -fanalyzer is still only suitable for analyzing C code. In particular, using it on C++ is unlikely to give meaningful output.

General Improvements

• For offload-device code generated via OpenMP and OpenACC, the math and the Fortran runtime libraries will now automatically be linked, when the user or compiler links them on the host side. Thus, it is no longer required to explicitly pass -lm and/or -lgfortran to the offload-device linker using the -foffload-options= flag.
• New configure options: --enable-host-pie, to build the compiler executables as PIE; and --enable-host-bind-now, to link the compiler executables with -Wl,-z,now in order to enable additional hardening.
• New option -fhardened, an umbrella option that enables a set of hardening flags. The options it enables can be displayed using the --help=hardened option.
• New option -fharden-control-flow-redundancy, to verify, at the end of functions, that the visited basic blocks correspond to a legitimate execution path, so as to detect and prevent attacks that transfer control into the middle of functions.
• New type attribute hardbool, for C and Ada. Hardened booleans take user-specified representations for true and false, presumably with higher hamming distance than standard booleans, and get verified at every use, detecting memory corruption and some malicious attacks.
• New type attribute strub to control stack scrubbing properties of functions and variables. The stack frame used by functions marked with the attribute gets zeroed-out upon returning or exception escaping. Scalar variables marked with the attribute cause functions contaning or accessing them to get stack scrubbing enabled implicitly.
• New option -finline-stringops, to force inline expansion of memcmp, memcpy, memmove and memset, even when that is not an optimization, to avoid relying on library implementations.
• New function attribute null_terminated_string_arg(PARAM_IDX) for indicating parameters that are expected to be null-terminated strings.
• The vectorizer now supports vectorizing loops which contain any number of early breaks. This means loops such as:

  	    int z[100], y[100], x[100];
  	    int foo (int n)
  	    {
  	      int res = 0;
  	      for (int i = 0; i < n; i++)
  		{
  		   y[i] = x[i] * 2;
  		   res += x[i] + y[i];
  
  		   if (x[i] > 5)
  		     break;
  
  		   if (z[i] > 5)
  		     break;
  
  		}
  	      return res;
  	    }
  	

  can now be vectorized on a number of targets. In this first version any input data sources must either have a statically known size at compile time or the vectorizer must be able to determine based on auxillary information that the accesses are aligned.

New Languages and Language specific improvements

• OpenMP
  • The GNU Offloading and Multi Processing Runtime Library Manual has been updated and extended, improving especially the description of ICVs, memory allocation, environment variables and OpenMP routines.
  • The requires directive's unified_address requirement is now fulfilled by both AMD GCN and nvptx devices. AMD GCN and nvptx devices now support low-latency allocators as detailed in the manual. Initial support for pinned-memory allocators has been added and, on Linux, libnuma is now used for allocators requesting the nearest-partition trait (both is described in the memory allocation section of the manual).
  • OpenMP 5.0: The allocate directive is now supported for stack variables in C and Fortran, including the OpenMP 5.1 align modifier. In C and C++ the map clause now accepts lvalue expressions. For Fortran, OpenMP allocators can now be used for allocatables and pointers using the allocate directive and its OpenMP 5.2 replacement, the allocators directive; files using this allocator and all files that might directly or indirectly (intrinisic assignment, intent(out), ...) de- or reallocate such-allocated variables must be compiled with the -fopenmp-allocators option.
  • OpenMP 5.1: Support was added for collapsing imperfectly nested loops and using present as map-type modifier and in defaultmap. The indirect clause is now supported. The performance of copying strided data from or to nvptx and AMD GPU devices using the OpenMP 5.1 routine omp_target_memcpy_rect has been improved.
  • OpenMP 5.2: The OMP_TARGET_OFFLOAD=mandatory handling has been updated for the clarifications and changes of the 5.2 specification. For Fortran, the list of directives permitted in Fortran pure procedures was extended. Additionally, the spec change has been implemented for default implicit mapping of C/C++ pointers pointing to unmapped storage. The destroy clause now optionally accepts the depend object as argument.
  • OpenMP 6.0 preview (TR11/TR12): The decl attribute is now supported in C++ 11 and the directive, sequence and decl attributes are now supported in C 23.
• OpenACC
  • OpenACC 2.7: The self clause was added to be used on compute constructs and the default clause for data constructs. Additionally, the readonly modifier is now handled in the copyin clause and cache directive.
  • OpenACC 3.2: The following API routines are now available in Fortran using the openacc module or the openacc_lib.h header file: acc_malloc, acc_free, acc_map_data, acc_unmap_data, acc_deviceptr, acc_hostptr, acc_memcpy_to_device, acc_memcpy_to_device_async, acc_memcpy_from_device, and acc_memcpy_from_device_async.

Ada

• New implementation-defined aspects and pragmas:
  • Local_Restrictions, which specifies that a particular subprogram does not violate one or more local restrictions, nor can it call a subprogram that is not subject to the same requirements.
  • User_Aspect_Definition and User_Aspect, which provide a mechanism for avoiding textual duplication if some set of aspect specifications is needed in multiple places.
• New implementation-defined aspects and pragmas for verification of the SPARK 2014 subset of Ada:
  • Always_Terminates, which provides a condition for a subprogram to necessarily complete (either return normally or raise an exception).
  • Ghost_Predicate, which introduces a subtype predicate that can reference Ghost entities.
  • Exceptional_Cases, which lists exceptions that might be propagated by the subprogram with side effects in the context of its precondition and associates them with a specific postcondition.
  • Side_Effects, which indicates that a function should be handled like a procedure with respect to parameter modes, Global contract, exceptional contract and termination: it may have output parameters, write global variables, raise exceptions and not terminate.
• The new attributes and contracts have been applied to the relevant parts of the Ada runtime library, which has been subsequently proven to be correct with SPARK 2014.
• Support for the LoongArch architecture.
• Support for vxWorks 7 Cert RTP has been removed.
• Additional hardening improvements. For more information reltated to hardening options, refer to the GCC Instrumentation Options and the GNAT Reference Manual, Security and Hardening Features.
• Improve style checking for redundant parentheses with -gnatyz
• New switch -gnateH to force reverse Bit_Order threshold to 64.
• Experimental features:
  • Storage Model: this feature proposes to redesign the concepts of Storage Pools into a more efficient model allowing higher performances and easier integration with low footprint embedded run-times.
  • String Interpolation: allows for easier string formatting.
• Further clean up and improvements to the GNAT code.

C family

• The Clang language extensions __has_feature and __has_extension have been implemented in GCC. These are available from C, C++, and Objective-C(++). This is primarily intended to aid the portability of code written against Clang.

C

• Some more C23 features have been implemented:
  • Bit-precise integer types (_BitInt (N) and unsigned _BitInt (N)): integer types with a specified number of bits. These are only supported on IA-32, x86-64 and AArch64 (little-endian) at present.
  • Structure, union and enumeration types may be defined more than once in the same scope with the same contents and the same tag; if such types are defined with the same contents and the same tag in different scopes, the types are compatible.
  • The <stdckdint.h> header for checked integer arithmetic.
• In addition to those C23 features, there are new command-line options -std=c23, -std=gnu23 and -Wc11-c23-compat. These are equivalent to the previous options -std=c2x, -std=gnu2x and -Wc11-c2x-compat, which are deprecated but remain supported.
• GCC supports a new pragma #pragma GCC novector to indicate to the vectorizer not to vectorize the loop annotated with the pragma.

C++

• Several C++26 features have been implemented:
  • P1854R4, Making non-encodable string literals ill-formed (PR110341)
  • P2752R3, Static storage for braced initializers (PR110346)
  • P2361R6, Unevaluated strings (PR110342)
  • P2738R1, constexpr cast from void* (PR110344)
  • P2741R3, User-generated static_assert messages (PR110348)
  • P2169R4, Placeholder variables with no name (PR110349)
  • P2864R2, Removing deprecated arithmetic conversion on enumerations
  • P2748R5, Disallow binding a returned reference to a temporary (PR114455)
  • P2809R3, Trivial infinite loops are not undefined behavior (PR114462)
• Several C++23 features have been implemented:
  • P0847R7, Deducing this (PR102609)
  • P2280R4, Using unknown references in constant expressions (PR106650)
  • P2564R3, consteval needs to propagate up (PR107687)
  • P2582R1, Class template argument deduction from inherited constructors (PR106653)
• Several C++ Defect Reports have been resolved, e.g.:
  • DR 532, Member/nonmember operator template partial ordering
  • DR 976, Deduction for const T& conversion operators
  • DR 2262, Attributes for asm-definition
  • DR 2359, Unintended copy initialization with designated initializers
  • DR 2386, tuple_size requirements for structured binding
  • DR 2406, [[fallthrough]] attribute and iteration statements
  • DR 2543, constinit and optimized dynamic initialization
  • DR 2586, Explicit object parameter for assignment and comparison
  • DR 2735, List-initialization and conversions in overload resolution
  • DR 2799, Inheriting default constructors
• When a diagnostic occurrs involving a C++ template, GCC will now quote the source code of the context at which the template is instantiated ("required from here"), rather than just print filename and line/column numbers.
• New built-in __type_pack_element to speed up traits such as std::tuple_element (PR100157)
• goto can cross the initialization of a trivially initialized object with a non-trivial destructor (DR 2256)
• -Wdangling-reference false positives have been reduced. The warning does not warn about std::span-like classes; there is also a new attribute gnu::no_dangling to suppress the warning. See the manual for more info.
• noexcept(expr) is now mangled as per the Itanium ABI
• the named return value optimization can now be performed even for variables declared in an inner block of a function, see the test
• New -Wnrvo warning, to warn if the named return value optimization is not performed although it is allowed by [class.copy.elision]. See the manual for more info.
• The backing array for std::initializer_list has been made static, allowing combining multiple equivalent initializer-lists (git)
• New -Welaborated-enum-base warning, to warn if an additional enum-base is used in an elaborated-type-specifier
• Better #include hints for missing headers (PR110164)
• The arguments of a variable template-id are coerced earlier than before, so various problems are detected earlier (PR89442)
• -Wmissing-field-initializers is no longer emitted for empty classes (PR110064)
• The constexpr code now tracks lifetimes in constant evaluation; this change helps to detect bugs such as accessing a variable whose lifetime has ended (PR70331, PR96630, PR98675)
• Array destruction can now be devirtualized
• In-class member variable template partial specializations are now accepted (PR71954)
• Improved diagnostic for explicit conversion functions: when a conversion doesn't work out only because the conversion function necessary to do the conversion couldn't be used because it was marked explicit, explain that to the user (git)
• Corrected mangling of static/thread_local structured bindings at function/block scope (PR111069)
• [basic.scope.block]/2 violations are detected even in compound-stmt of function-try-block and for block-scope external variables (PR52953)
• Improved "not a constant expression" diagnostic when taking the address of a non-static constexpr variable (PR91483)
• Non-dependent simple assignments are checked even in templates (PR18474)
• Attributes hot and cold can be applied to classes as well. See the manual for more info.
• Function template constraints, as well as CTAD placeholders, are now mangled
• Various decltype fixes: PR79620, PR79378, PR83167, PR96917
• New option -fdiagnostics-all-candidates to note all candidates during overload resolution failure
• -Walloc-size and -Wcalloc-transposed-args warnings are enabled for C++ as well
• The DR 2237 code no longer gives an error, it emits a -Wtemplate-id-cdtor warning instead
• GCC supports a new pragma #pragma GCC novector to indicate to the vectorizer not to vectorize the loop annotated with the pragma.
• C++ module scanning for named modules is now available, based on the format described in P1689R5, Format for describing dependencies of source files. The -fdeps-format=, -fdeps-file=, and -fdeps-target= flags may be used to generate dependency information. In GCC 14 p1689r5 is the only valid argument for -fdeps-format=.

Runtime Library (libstdc++)

• The libstdc++exp.a library now includes all the Filesystem TS symbols from the libstdc++fs.a library. The experimental symbols for the C++23 std::stacktrace class are also in libstdc++exp.a, replacing the libstdc++_libbacktrace.a library that GCC 13 provides. This means that -lstdc++exp is the only library needed for all experimental libstdc++ features.
• Improved experimental support for C++20, including:
  • std::chrono::parse.
  • Unicode-aware string handling in std::format.
• Improved experimental support for C++23, including:
  • The std::ranges::to function for converting ranges to containers.
  • The std::generator view for getting results from coroutines.
  • The <stacktrace> header is supported by default.
  • std::print and std::println (requires linking with -lstdc++exp on Windows).
  • Formatters for std::thread::id and std::stacktrace.
  • Smart pointer adaptors, std::out_ptr and std::inout_ptr.
  • Some range adaptors now support move-only types.
• Experimental support for C++26, including:
  • Native handles for filebuf, fstream, etc.
  • Functions for saturation arithmetic on integers.
  • std::to_string now uses std::format.
  • Enhanced formatting of pointers with std::format.
  • The std::runtime_format function to allow using non-literal format strings with std::format.
  • Testable result types for <charconv> functions.
  • The std::text_encoding class for identifying character sets (requires linking with -lstdc++exp for some member functions).
• Faster numeric conversions using std::to_string and std::to_wstring.
• Updated parallel algorithms that are compatible with oneTBB.
• std::numeric_limits<_Float32> and std::numeric_limits<_Float64> are now defined for all standard modes, not only for C++23.
• Added missing functions for float and long double to <cmath>.
• Using the std::setfill manipulator with std::istream is deprecated.

Fortran

• The compiler now accepts the -std=f2023 option, which has been added in preparation of support of Fortran 2023. This option increases the line-length limit for source in free-form to 10000, and statements may have up to 1 million characters.
• With the -save-temps option, preprocessed files with the .fii extension will be generated from free-form source files such as .F90 and .fi from fixed-form files such as .F.

Modula-2

• The automatic dependency generation options: -M, -MD, -MF, -MMD, -MP, -MQ and -MT have been implemented in the compiler.
• The -Wcase-enum and -Wuninit-variable-checking= options have been implemented to provide compile time warnings against missing case clauses and uninitialized variables respectively.

libgccjit

• The libgccjit API gained 6 new entry points:
  • gcc_jit_type_get_restrict for adding restrict to types (LIBGCCJIT_ABI_25).
  • 4 functions for setting attributes on functions and variables (LIBGCCJIT_ABI_26):
    • gcc_jit_function_add_attribute
    • gcc_jit_function_add_string_attribute
    • gcc_jit_function_add_integer_array_attribute
    • gcc_jit_lvalue_add_string_attribute
  • gcc_jit_context_new_sizeof for accessing the size of a type (LIBGCCJIT_ABI_27).

New Targets and Target Specific Improvements

AArch64

• A number of new CPUs are supported through the -mcpu and -mtune options (GCC identifiers in parentheses).
  • Ampere-1B (ampere1b).
  • Arm Cortex-A520 (cortex-a520).
  • Arm Cortex-A720 (cortex-a720).
  • Arm Cortex-X4 (cortex-x4).
  • Microsoft Cobalt-100 (cobalt-100).
• Additionally, the identifiers generic, generic-armv8-a and generic-armv9-a are added as arguments to -mcpu= and -mtune= to optimize code generation aimed at a good blend of CPUs of a particular architecture version. These tunings are also used as the default optimization targets when compiling with the -march=armv8-a or -march=armv9-a options and their point releases e.g. -march=armv8.2-a or -march=armv9.3-a.
• New features in the Arm architecture are supported in a number of ways:
  • Support is added for the Arm Streaming Matrix Extensions SME and SME2 through the +sme and +sme2 extensions to -march=. In particular, this includes support for the Beta state of the SME ACLE in the form of a new intrinsics arm_sme.h intrinsics header and a number of new keyword attributes to manage use of the new Streaming SVE state. For more information please refer to the ACLE documentation.
  • Libatomic is updated to implement 128-bit atomic operations locklessly on systems with FEAT_LSE2.
  • Support for FEAT_LRCPC3 is added through ACLE intrinsics in arm_neon.h header and enabled through the +rcpc3 extension to -march=.
• As well as numerous AArch64 code generation improvements, the following optimization enhancements are noteworthy:
  • A new AArch64-specific register allocation pass is added. It runs in addition to standard register allocation. The pass's main purpose is to make use of strided vector register operands in SME instructions. However, it can also remove redundant moves in normal Advanced SIMD and SVE code. The pass is controlled by the new option -mearly-ra= that takes the arguments all, strided, none. -mearly-ra=all is enabled by default at optimization levels -O2 and above.
  • A new optimization pass to fuse loads and stores to adjacent memory locations into load and store-pair AArch64 instructions. The pass is enabled by default when compiling with optimization and runs twice in the optimization pipeline: before and after register allocation. This can be controlled with the options -mearly-ldp-fusion and -mlate-ldp-fusion.
• Conformance with the ACLE specification is improved and a number of features aimed at helping developers deploy Arm architecture features are added:
  • Support for the Beta version of the Function Multiversioning Specification. This feature provides facilities to annotate functions with attributes that allow the compiler to generate multiple versions of the function, selected at runtime based on the architecture features available in the system. Please refer to the ACLE specification for more details.
  • Support for more ACLE intrinsics in the arm_acle.h header, including the Memory prefetch intrinsics and the Special register intrinsics. This also includes intrinsics for the extension to 128-bit system registers, enabled through the +d128 extension to -march=.
  • Intrinsics enabled by the +dotprod, +fp16, +fp16fml, +i8mm, +sha3 and +sm4 extensions to -march= no longer require -march=armv8.2-a or higher to be specified. Likewise, the intrinsics enabled by +memtag no longer require -march=armv8.5-a.
  • Support for the NEON-SVE Bridge intrinsics. These are intrinsics that allow conversions between NEON and SVE vectors, enabled through the inclusion of the arm_neon_sve_bridge.h header.
• The option -mtp= is now supported for changing the TPIDR register used for TLS accesses. For more details please refer to the documentation.

AMD Radeon (GCN)

• Initial support for the AMD Radeon gfx90c (GCN5), gfx1030, gfx1036 (RDNA2), gfx1100 and gfx1103 (RDNA3) devices has been added. LLVM 15+ (assembler and linker) is required to support GFX11.
• Improved register usage and performance on CDNA Instinct MI100 and MI200 series devices.
• The default device architecture is now gfx900 (Vega).
• Fiji (gfx803) device support is now deprecated and will be removed from a future release. The default compiler configuration no longer uses Fiji as the default device, and no longer includes the Fiji libraries. Both can be restored by configuring with --with-arch=fiji .

arm

• The Cortex-M52 CPU is now supported through the cortex-m52 argument to the -mcpu and -mtune options.

AVR

• On AVR64* and AVR128* devices, read-only data is now located in program memory per default and no longer in RAM.
  • Only a 32 KiB block of program memory can be used to store and access .rodata in that way. Which block is used can be selected by defining the symbol __flmap. As an alternative, the byte address of the block can be specified by the symbol __RODATA_FLASH_START__ which takes precedence over __flmap. For example, linking with -Wl,--defsym,__RODATA_FLASH_START__=32k chooses the second 32 KiB block.
  • The default uses the last 32 KiB block, which is also the hardware default for bit-field NVMCTRL_CTRLB.FLMAP.
  • When a non-default block is used, then NVMCTRL_CTRLB.FLMAP must be initialized accordingly by hand, or AVR-LibC v2.2 that implements #931 can be used. The latter initializes NVMCTRL_CTRLB.FLMAP in the startup code and according to the value of __flmap or __RODATA_FLASH_START__.
  • When AVR-LibC with #931 is used, then defining the symbol __flmap_lock to a non-zero value will set bit NVMCTRL_CTRLB.FLMAPLOCK. This will protect NVMCTRL_CTRLB.FLMAP from any further changes — which would be Undefined Behaviour in C/C++. If you prefer to define the symbol in a C/C++ file, an asm statement can be used:

    __asm (".global __flmap_lock"  "\n\t"
           "__flmap_lock = 1");

  • When you do not want the code from #931, then define a global symbol __do_flmap_init and the linker will not pull in that code from libmcu.a any more.
  • In order to return to the old placement of read-only data in RAM, the new compiler option -mrodata-in-ram can be used. This is required on devices where the hardware revision is affected by a silicon bug concerning the FLMAP functionality.
  • Read-only data is located in output section .rodata, whereas it is part of .text when located in RAM.
  • The feature is only available when the compiler is configured with a version of Binutils that implements PR31124, which is the case for Binutils v2.42 and up.
  • The implementation consists of two parts:
    1. Binutils support new emulations avrxmega2_flmap and avrxmega4_flmap. The sole purpose of these emulations is to provide adjusted default linker description files. Apart from that, these emulations behave exactly the same like avrxmega2 resp. avrxmega4.
    2. The compiler uses a device-specs file which links the program with -mavrxmega2_flmap or -mavrxmega2 depending on -m[no-]rodata-in-ram; and similar for -mavrxmega4[_flmap].
    This means the feature can be used with older compiler or Binutils versions; all what's needed is an adjusted linker script and a custom device-specs file.
• A new compiler option -m[no-]rodata-in-ram has been added. The default is to locate read-only data in program memory for devices that support it, e.g. for AVR64* and AVR128* devices as explained above, and for devices from the avrxmega3 and avrtiny families.
• The new built-in macro __AVR_RODATA_IN_RAM__ is supported on all devices. It's defined to 0 or 1.
• A new optimization tries to improve code generation for indirect memory accesses on Reduced Tiny devices. It can be controlled by the new compiler option -mfuse-add=level where level may be 0, 1 or 2.
• On the Reduced Tiny devices, the meaning of register constraint "w" has been changed. It now constrains the registers R24…R31 as is the case for all the other devices.

IA-32/x86-64

• New compiler option -m[no-]evex512 was added. The compiler switch enables/disables 512-bit vector. It will be default on if AVX512F is enabled.
• Part of new feature support for Intel APX was added, including EGPR, NDD, PPX and PUSH2POP2. APX support is available via the -mapxf compiler switch.
• For inline asm support with APX, by default the EGPR feature was disabled to prevent potential illegal instruction with EGPR occurs. To invoke egpr usage in inline asm, use new compiler option -mapx-inline-asm-use-gpr32 and user should ensure the instruction supports EGPR.
• New ISA extension support for Intel AVX10.1 was added. AVX10.1 intrinsics are available via the -mavx10.1 or -mavx10.1-256 compiler switch with 256-bit vector size support. 512-bit vector size support for AVX10.1 intrinsics are available via the -mavx10.1-512 compiler switch.
• New ISA extension support for Intel AVX-VNNI-INT16 was added. AVX-VNNI-INT16 intrinsics are available via the -mavxvnniint16 compiler switch.
• New ISA extension support for Intel SHA512 was added. SHA512 intrinsics are available via the -msha512 compiler switch.
• New ISA extension support for Intel SM3 was added. SM3 intrinsics are available via the -msm3 compiler switch.
• New ISA extension support for Intel SM4 was added. SM4 intrinsics are available via the -msm4 compiler switch.
• New ISA extension support for Intel USER_MSR was added. USER_MSR intrinsics are available via the -muser_msr compiler switch.
• GCC now supports the Intel CPU named Clearwater Forest through -march=clearwaterforest. Based on Sierra Forest, the switch further enables the AVX-VNNI-INT16, PREFETCHI, SHA512, SM3, SM4 and USER_MSR ISA extensions.
• GCC now supports the Intel CPU named Arrow Lake through -march=arrowlake. Based on Alder Lake, the switch further enables the AVX-IFMA, AVX-NE-CONVERT, AVX-VNNI-INT8 and CMPccXADD ISA extensions.
• GCC now supports the Intel CPU named Arrow Lake S through -march=arrowlake-s. Based on Arrow Lake, the switch further enables the AVX-VNNI-INT16, SHA512, SM3 and SM4 ISA extensions.
• GCC now supports the Intel CPU named Lunar Lake through -march=lunarlake. Lunar Lake is based on Arrow Lake S.
• GCC now supports the Intel CPU named Panther Lake through -march=pantherlake. Based on Arrow Lake S, the switch further enables the PREFETCHI ISA extensions.
• Xeon Phi CPUs support (a.k.a. Knight Landing and Knight Mill) are marked as deprecated. GCC will emit a warning when using the -mavx5124fmaps, -mavx5124vnniw, -mavx512er, -mavx512pf, -mprefetchwt1, -march=knl, -march=knm, -mtune=knl or -mtune=knm compiler switches. Support will be removed in GCC 15.
• Hardware-assisted AddressSanitizer now works for the x86-64 target with LAM_U57. -fsanitize=hwaddress will enable -mlam=u57 by default.
• GCC now supports AMD CPUs based on the znver5 core via -march=znver5. In addition to the ISA extensions enabled on a znver4 core, this switch further enables the AVX512VP2INTERSECT, AVXVNNI, MOVDIR64B, MOVDIRI, and PREFETCHI ISA extensions.

MCore

• Bitfields are now signed by default per GCC policy. If you need bitfields to be unsigned, use -funsigned-bitfields.

LoongArch

• Support for the following -march parameters has been added:
  • la64v1.0
  • la64v1.1
  • la664
  It is now recommended to use -march=la64v1.0 as the only compiler option to describe the target ISA when building binaries for distribution. For more information on LoongArch ISA versions, see Toolchain Conventions of the LoongArch™ Architecture.
• Support for the following -mtune parameters has been added:
  • generic
  • la664
• New ISA Extension
  • LSX (Loongson SIMD Extension): Support 128-bit vector instructions and the intrinsics.
  • LASX (Loongson Advanced SIMD Extension): Support 256-bit vector instructions and the intrinsics.
  • FRECIPE: Support frecipe.{s/d} and frsqrte.{s/d} instructions and the intrinsics.
  • DIV32: Support div.w[u] and mod.w[u] instructions with inputs not sign-extended.
  • LAM_BH: Support am{swap/add}[_db].{b/h} instructions.
  • LAMCAS: Support amcas[_db].{b/h/w/d} instructions.
• New Built-in Macros
  • __loongarch_arch: Target ISA preset as specified by -march=. For example, compiling with -march=la64v1.0, the value of __loongarch_arch is "la64v1.0".
  • __loongarch_tune: Processor model as specified by -mtune or its default value.
  • __loongarch_{simd,sx,asx}: These macros are not defined, or defined as 1.
  • __loongarch_simd_width: The maximum SIMD bit-width enabled by the compiler. (128 for lsx, and 256 for lasx).
  • __loongarch_frecipe: It's defined to 1 or undefined.
  • __loongarch_div32: It's defined to 1 or undefined.
  • __loongarch_lam_bh: It's defined to 1 or undefined.
  • __loongarch_lamcas: It's defined to 1 or undefined.
  • __loongarch_ld_seq_sa: It's defined to 1 or undefined.
  • __loongarch_version_major: The minimally required LoongArch ISA version (major) to run the compiled program, defined to 1 or undefined (iff no such version is known to the compiler).
  • __loongarch_version_minor: The minimally required LoongArch ISA version (minor) to run the compiled program, defined to 0 1 or undefined (iff __loongarch_version_major is undefined).
  • __FLOAT128_TYPE: It's defined to 1.
• New Intrinsics
  • __builtin_thread_pointer
  • __lsx_*
  • __lasx_*
  • __frecipe_{s/d} and __frsqrte_{s/d}
• New Compiler Option
  • -m[no-]lsx
  • -m[no-]lasx
  • -m[no-]frecipe
  • -m[no-]div32
  • -m[no-]lam-bh
  • -m[no-]lamcas
  • -m[no-]ld-seq-sa
  • -mrecip=
  • -m[no-]recip
  • -mexplicit-relocs={none,always,auto}
  • -m[no-]relax
  • -m[no-]pass-mrelax-to-as
  • -mtls-dialect={trad,desc}
• Support for Ada and D.
• Support for libffi.
• Enable -free by default at -O2 or higher.
• Enable -fsched-pressure by default at -O1 or higher.
• Support the extreme code model using macro instructions (under -mno-explicit-relocs).
• Support call36.
• Optimizing built-in functions for memory-model-aware atomic operations using hierarchical dbar instructions.
• TLS descriptors support. It is not enabled by default, and can be enabled with -mtls-dialect=desc. The default behavior can be configured with --with-tls=[trad|desc].

RISC-V

• The SLP and loop vectorizer are now enabled for RISC-V when the vector extension is enabled, thanks to Ju-Zhe Zhong from RiVAI, Pan Li from Intel, and Robin Dapp from Ventana Micro for contributing most of the implementation!
• The -mrvv-max-lmul= option has been introduced for performance tuning of the loop vectorizer. The default value is -mrvv-max-lmul=m1, which limits the maximum LMUL to 1. The -mrvv-max-lmul=dynamic setting can dynamically select the maximum LMUL value based on register pressure.
• Atomic code generation has been improved and is now in conformance with the latest psABI specification, thanks to Patrick O'Neill from Rivos.
• Support for the vector intrinsics as specified in version 1.0 of the RISC-V vector intrinsic specification.
• Support for the experimental vector crypto intrinsics as specified in RISC-V vector intrinsic specification, thanks to Feng Wang et al. from ESWIN Computing
• Support for the T-head vector intrinsics.
• Support for the scalar bitmanip and scalar crypto intrinsics, thanks to Liao Shihua from PLCT.
• Support for the large code model via option -mcmodel=large, thanks to Kuan-Lin Chen from Andes Technology.
• Support for the standard vector calling convention variant, thanks to Lehua Ding from RiVAI.
• Supports the target attribute, which allows users to compile a function with specific extensions.
• -march= option no longer requires the architecture string to be in canonical order, with only a few constraints remaining: the architecture string must start with rv[32|64][i|g|e], and must use an underscore as the separator after a multi-letter extension.
• -march=help option has been introduced to dump all supported extensions.
• Added experimental support for the -mrvv-vector-bits=zvl option and the riscv_rvv_vector_bits attribute, which specify a fixed length for scalable vector types. This option is optimized for specific vector core implementations; however, the code generated with this option is NOT portable between the core with different VLEN, thanks to Pan Li from Intel.
• Support for TLS descriptors has been introduced, which can be enabled by the -mtls-dialect=desc option. The default behavior can be configured with --with-tls=[trad|desc].
• Support for the TLS descriptors, this can be enabled by -mtls-dialect=desc and the default behavior can be configure by --with-tls=[trad|desc], and this feature require glibc 2.40, thanks to Tatsuyuki Ishi from Blue Whale Systems
• Support for the following standard extensions has been added:
  • Vector crypto extensions:
    • Zvbb
    • Zvkb
    • Zvbc
    • Zvkg
    • Zvkned
    • Zvkhna
    • Zvkhnb
    • Zvksed
    • Zvksh
    • Zvkn
    • Zvknc
    • Zvkng
    • Zvks
    • Zvksc
    • Zvksg
    • Zvkt
  • Code size reduction extensions:
    • Zca
    • Zcb
    • Zce
    • Zcf
    • Zcd
    • Zcmp
    • Zcmt
  • Zicond
  • Zfa
  • Ztso
  • Zvfbfmin
  • Zvfhmin
  • Zvfh
  • Za64rs
  • Za128rs
  • Ziccif
  • Ziccrse
  • Ziccamoa
  • Zicclsm
  • Zic64b
  • Smaia
  • Smepmp
  • Smstateen
  • Ssaia
  • Sscofpmf
  • Ssstateen
  • Sstc
  • Svinval
  • Svnapot
  • Svpbmt
• Support for the following vendor extensions has been added:
  • T-Head:
    • XTheadVector
  • CORE-V:
    • XCVmac
    • XCValu
    • XCVelw
    • XCVsimd
    • XCVbi
  • Ventana Micro:
    • XVentanaCondops
• The following new CPUs are supported through the -mcpu option (GCC identifiers in parentheses).
  • SiFive's X280 (sifive-x280).
  • SiFive's P450 (sifive-p450).
  • SiFive's P670 (sifive-p670).
• The following new CPUs are supported through the -mtune option (GCC identifiers in parentheses).
  • Generic out-of-order core (generic-ooo).
  • SiFive's P400 series (sifive-p400-series).
  • SiFive's P600 series (sifive-p600-series).
  • XiangShan's Nanhu microarchitecture (xiangshan-nanhu).

SPARC

• The implementation of calling conventions for small structures containing arrays of floating-point components has been changed in 64-bit mode for the Solaris port to match the implementation of the vendor compiler (and the ABI). As a result, the code generated will not be binary compatible with earlier releases in these cases.

Documentation improvements

• GCC's ability to provide clickable hyperlinks to the documentation has been extended, so that whenever GCC refers to a command-line option in quotes in a diagnostic message, the option is a clickable hyperlink (assuming a suitably capable terminal).

Improvements to Static Analyzer

• New warnings:
  • -Wanalyzer-infinite-loop warns about paths through the code which appear to lead to an infinite loop.
  • -Wanalyzer-overlapping-buffers warns for paths through the code in which overlapping buffers are passed to an API for which the behavior on such buffers is undefined.
  • -Wanalyzer-undefined-behavior-strtok warns for paths through the code in which a call is made to strtok with undefined behavior.
• Previously, the analyzer's "taint" tracking to be explicitly enabled via -fanalyzer-checker=taint (along with -fanalyzer). This is now enabled by default when -fanalyzer is selected, thus also enabling the 6 taint-based warnings:
  • -Wanalyzer-tainted-allocation-size
  • -Wanalyzer-tainted-array-index
  • -Wanalyzer-tainted-assertion
  • -Wanalyzer-tainted-divisor
  • -Wanalyzer-tainted-offset
  • -Wanalyzer-tainted-size
• The analyzer will now simulate API calls that expect null-terminated string arguments, and will warn about code paths in which such a call is made with a buffer that isn't properly terminated, either due to a read of an uninitialized byte or an out-of-range accesses seen before any zero byte is seen. This applies to functions that use the new null_terminated_string_arg(PARAM_IDX) attribute, functions that use the format attribute, and to the library functions error (parameter 3), error_at_line (parameter 5), putenv, strchr (parameter 1), and strcpy (parameter 2).
• The analyzer now makes use of the function attribute alloc_size allowing -fanalyzer to emit -Wanalyzer-allocation-size, -Wanalyzer-out-of-bounds, and -Wanalyzer-tainted-allocation-size on execution paths involving allocations using such functions.
• The analyzer's knowledge about the behavior of the standard library has been extended to cover fopen, strcat, strncpy, and strstr. The analyzer will also more precisely model the behavior of memcpy, memmove, strcpy, strdup, strlen, and of various atomic built-in functions.
• The warning -Wanalyzer-out-of-bounds has been extended so that, where possible, it will emit a text-based diagram visualizing the spatial relationship between
  1. the memory region that the analyzer predicts would be accessed, versus
  2. the range of memory that is valid to access
  whether they overlap, are touching, are close or far apart; which one is before or after in memory, the relative sizes involved, the direction of the access (read vs write), and, in some cases, the values of data involved.

  Such "text art" diagrams can be controlled (or suppressed) via a new -fdiagnostics-text-art-charset= option.

  For example, given the out-of-bounds write in strcat in:

  void test (void)
  {
     char buf[10];
     strcpy (buf, "hello");
     strcat (buf, " world!");
  }
  

  it emits:

                                   ┌────┬────┬────┬────┬────┐┌─────┬─────┬─────┐
                                   │[0] │[1] │[2] │[3] │[4] ││ [5] │ [6] │ [7] │
                                   ├────┼────┼────┼────┼────┤├─────┼─────┼─────┤
                                   │' ' │'w' │'o' │'r' │'l' ││ 'd' │ '!' │ NUL │
                                   ├────┴────┴────┴────┴────┴┴─────┴─────┴─────┤
                                   │     string literal (type: 'char[8]')      │
                                   └───────────────────────────────────────────┘
                                     │    │    │    │    │      │     │     │
                                     │    │    │    │    │      │     │     │
                                     v    v    v    v    v      v     v     v
        ┌─────┬────────────────────┬────┬──────────────┬────┐┌─────────────────┐
        │ [0] │        ...         │[5] │     ...      │[9] ││                 │
        ├─────┼────┬────┬────┬────┬┼────┼──────────────┴────┘│                 │
        │ 'h' │'e' │'l' │'l' │'o' ││NUL │                    │after valid range│
        ├─────┴────┴────┴────┴────┴┴────┴───────────────────┐│                 │
        │             'buf' (type: 'char[10]')              ││                 │
        └───────────────────────────────────────────────────┘└─────────────────┘
        ├─────────────────────────┬─────────────────────────┤├────────┬────────┤
                                  │                                   │
                        ╭─────────┴────────╮                ╭─────────┴─────────╮
                        │capacity: 10 bytes│                │overflow of 3 bytes│
                        ╰──────────────────╯                ╰───────────────────╯
  

  showing that the overflow occurs partway through the second string fragment.
• The analyzer will now attempt to track execution paths involving computed gotos, whereas previously it gave up on such paths.

Improvements to SARIF support

• The SARIF output from -fdiagnostics-format= now adds indentation and newlines to reflect the logical JSON structure of the data. The previous compact behavior can be restored via the new option -fno-diagnostics-json-formatting. This also applies to the older output format named "json".
• If profiling information about the compiler itself is requested via -ftime-report, and a SARIF output format is requested via -fdiagnostics-format=, then the timing and memory usage data is now written in JSON form into the SARIF output, rather than as plain text to stderr.

Improvements for plugin authors

• GCC diagnostics have been able to have execution paths associated with them since GCC 10, but previously these were required to be single-threaded. As of GCC 14, these execution paths can have multipled named threads associated with them, with each event being associated with one of the threads. No existing GCC diagnostics take advantage of this, but GCC plugins may find this useful for their own diagnostics; an example is provided in the testsuite.
• GCC's diagnostics can now optionally add per-diagnostic property bags to the SARIF output, allowing plugins to capture custom data as needed with their diagnostics.

GCC 14.1

This is the list of problem reports (PRs) from GCC's bug tracking system that are known to be fixed in the 14.1 release. This list might not be complete (that is, it is possible that some PRs that have been fixed are not listed here).