1============================ 2LLVM |release| Release Notes 3============================ 4 5.. contents:: 6 :local: 7 8 9Introduction 10============ 11 12This document contains the release notes for the LLVM Compiler Infrastructure, 13release |release|. Here we describe the status of LLVM, including major improvements 14from the previous release, improvements in various subprojects of LLVM, and 15some of the current users of the code. All LLVM releases may be downloaded 16from the `LLVM releases web site <https://llvm.org/releases/>`_. 17 18For more information about LLVM, including information about the latest 19release, please check out the `main LLVM web site <https://llvm.org/>`_. If you 20have questions or comments, the `LLVM Developer's Mailing List 21<https://lists.llvm.org/mailman/listinfo/llvm-dev>`_ is a good place to send 22them. 23 24Note that if you are reading this file from a Git checkout or the main 25LLVM web page, this document applies to the *next* release, not the current 26one. To see the release notes for a specific release, please see the `releases 27page <https://llvm.org/releases/>`_. 28 29Non-comprehensive list of changes in this release 30================================================= 31.. NOTE 32 For small 1-3 sentence descriptions, just add an entry at the end of 33 this list. If your description won't fit comfortably in one bullet 34 point (e.g. maybe you would like to give an example of the 35 functionality, or simply have a lot to talk about), see the `NOTE` below 36 for adding a new subsection. 37 38* ... 39 40Update on required toolchains to build LLVM 41------------------------------------------- 42 43With LLVM 15.x we will raise the version requirements of the toolchain used 44to build LLVM. The new requirements are as follows: 45 46* GCC >= 7.1 47* Clang >= 5.0 48* Apple Clang >= 9.3 49* Visual Studio 2019 >= 16.7 50 51In LLVM 15.x these requirements will be "soft" requirements and the version 52check can be skipped by passing ``-DLLVM_TEMPORARILY_ALLOW_OLD_TOOLCHAIN=ON`` 53to CMake. 54 55With the release of LLVM 16.x these requirements will be hard and LLVM developers 56can start using C++17 features, making it impossible to build with older 57versions of these toolchains. 58 59Changes to the LLVM IR 60---------------------- 61 62* LLVM now uses `opaque pointers <OpaquePointers.html>`__. This means that 63 different pointer types like ``i8*``, ``i32*`` or ``void()**`` are now 64 represented as a single ``ptr`` type. See the linked document for migration 65 instructions. 66* Renamed ``llvm.experimental.vector.extract`` intrinsic to ``llvm.vector.extract``. 67* Renamed ``llvm.experimental.vector.insert`` intrinsic to ``llvm.vector.insert``. 68* The constant expression variants of the following instructions have been 69 removed: 70 71 * ``extractvalue`` 72 * ``insertvalue`` 73 * ``udiv`` 74 * ``sdiv`` 75 * ``urem`` 76 * ``srem`` 77 * ``fadd`` 78 * ``fsub`` 79 * ``fmul`` 80 * ``fdiv`` 81 * ``frem`` 82 83* Added the support for ``fmax`` and ``fmin`` in ``atomicrmw`` instruction. The 84 comparison is expected to match the behavior of ``llvm.maxnum.*`` and 85 ``llvm.minnum.*`` respectively. 86* ``callbr`` instructions no longer use ``blockaddress`` arguments for labels. 87 Instead, label constraints starting with ``!`` refer directly to entries in 88 the ``callbr`` indirect destination list. 89 90.. code-block:: llvm 91 92 ; Old representation 93 %res = callbr i32 asm "", "=r,r,i"(i32 %x, i8 *blockaddress(@foo, %indirect)) 94 to label %fallthrough [label %indirect] 95 ; New representation 96 %res = callbr i32 asm "", "=r,r,!i"(i32 %x) 97 to label %fallthrough [label %indirect] 98 99Changes to building LLVM 100------------------------ 101 102* Omitting ``CMAKE_BUILD_TYPE`` when using a single configuration generator is now 103 an error. You now have to pass ``-DCMAKE_BUILD_TYPE=<type>`` in order to configure 104 LLVM. This is done to help new users of LLVM select the correct type: since building 105 LLVM in Debug mode is very resource intensive, we want to make sure that new users 106 make the choice that lines up with their usage. We have also improved documentation 107 around this setting that should help new users. You can find this documentation 108 `here <https://llvm.org/docs/CMake.html#cmake-build-type>`_. 109 110Changes to TableGen 111------------------- 112 113Changes to Loop Optimizations 114----------------------------- 115 116* Loop interchange legality and cost model improvements 117 118 119Changes to the AArch64 Backend 120------------------------------ 121 122Changes to the AMDGPU Backend 123----------------------------- 124 125* 8 and 16-bit atomic loads and stores are now supported 126 127 128Changes to the ARM Backend 129-------------------------- 130 131* Added support for the Armv9-A, Armv9.1-A and Armv9.2-A architectures. 132* Added support for the Armv8.1-M PACBTI-M extension. 133* Added support for the Armv9-A, Armv9.1-A and Armv9.2-A architectures. 134* Added support for the Armv8.1-M PACBTI-M extension. 135* Removed the deprecation of ARMv8-A T32 Complex IT blocks. No deprecation 136 warnings will be generated and -mrestrict-it is now always off by default. 137 Previously it was on by default for Armv8 and off for all other architecture 138 versions. 139* Added a pass to workaround Cortex-A57 Erratum 1742098 and Cortex-A72 140 Erratum 1655431. This is enabled by default when targeting either CPU. 141* Implemented generation of Windows SEH unwind information. 142* Switched the MinGW target to use SEH instead of DWARF for unwind information. 143* Added support for the Cortex-M85 CPU. 144* Added support for a new ``-mframe-chain=(none|aapcs|aapcs+leaf)`` command-line 145 option, which controls the generation of AAPCS-compliant Frame Records. 146 147Changes to the AVR Backend 148-------------------------- 149 150* ... 151 152Changes to the DirectX Backend 153------------------------------ 154 155* DirectX has been added as an experimental target. Specify 156 ``-DLLVM_EXPERIMENTAL_TARGETS_TO_BUILD=DirectX`` in your CMake configuration 157 to enable it. The target is not packaged in pre-built binaries. 158* The DirectX backend supports the ``dxil`` architecture which is based on LLVM 159 3.6 IR encoded as bitcode and is the format used for DirectX GPU Shader 160 programs. 161 162Changes to the Hexagon Backend 163------------------------------ 164 165* ... 166 167Changes to the MIPS Backend 168--------------------------- 169 170* ... 171 172Changes to the PowerPC Backend 173------------------------------ 174 175Common PowerPC improvements: 176* Add a new post instruction selection pass to generate CTR loops. 177* Add SSE4 and BMI compatible intrinsics implementation. 178* Supported 16-byte lock free atomics on PowerPC8 and up. 179* Supported atomic load/store for pointer types. 180* Supported stack size larger than 2G 181* Add __builtin_min/__builtin_max/__abs builtins. 182* Code generation improvements for splat load/vector shuffle/mulli, etc. 183* Emit VSX instructions for vector loads and stores regardless of alignment. 184* The mcpu=future has its own ISA now (FutureISA). 185* Added the ppc-set-dscr option to set the Data Stream Control Register (DSCR). 186* Bug fixes. 187 188AIX improvements: 189* Supported 64 bit XCOFF for integrated-as path. 190* Supported X86-compatible vector intrinsics. 191* Program code csect default alignment now is 32-byte. 192* Supported auxiliary header in integrated-as path. 193* Improved alias symbol handling. 194 195Changes to the RISC-V Backend 196----------------------------- 197 198* A RISCVRedundantCopyElimination pass was added to remove unnecessary zero 199 copies. 200* A RISC-V specific CodeGenPrepare pass was added. 201* The machine outliner was enabled by default for RISC-V at ``-Oz``. 202 Additionally, the newly introduced RISCVMakeCompressible pass will make 203 modify instructions prior to emission at ``-Oz`` in order to increase 204 opportunities for the compression with the RISC-V C extension. 205* Various bug fixes and improvements to code generation for the RISC-V vector 206 extensions. 207* Various improvements were made to RISC-V specific optimisation passes such 208 as RISCVSExtWRemoval and RISCVMergeBaseOffset. 209* llc now computes the target ABI based on the target architecture using the 210 same logic as Clang if not explicit ABI is given. 211* ``generic`` is now recognized as a valid CPU name and is mapped to 212 ``generic-rv32`` or ``generic-rv64`` depending on the target triple. 213* Support for the experimental Zvfh extension was added, enabling 214 half-precision floating point in vectors. 215* Support for the Zihintpause (Pause Hint) extension. 216* Assembler and disassembler support for the Zfinx and Zdinx (float / double 217 in integer register) extensions. 218* Assembler and disassembler support for the Zicbom, Zicboz, and Zicbop cache 219 management operation extensions. 220* Support for the Zmmul extension (a subextension of the M extension, adding 221 multiplication instructions only). 222* Assembler and disassembler support for the hypervisor extension and for the 223 Sinval supervisor memory-management extension. 224 225Changes to the WebAssembly Backend 226---------------------------------- 227 228* ... 229 230Changes to the SystemZ Backend 231------------------------------ 232 233* Support z16 processor name. 234* Machine scheduler description for z16. 235* Add support for inline assembly address operands ("p") as well as for SystemZ 236 specific address operands ("ZQ", "ZR", "ZS" and "ZT"). 237* Efficient handling of small memcpy/memset operations up to 32 bytes. 238* Tuning of the inliner. 239* Fixing emission of library calls so that narrow integer arguments are sign or 240 zero extended per the SystemZ ABI. 241* Support added for libunwind. 242* Various minor improvements and bugfixes. 243 244Changes to the X86 Backend 245-------------------------- 246 247* Support ``half`` type on SSE2 and above targets following X86 psABI. 248* Support ``rdpru`` instruction on Zen2 and above targets. 249 250During this release, ``half`` type has an ABI breaking change to provide the 251support for the ABI of ``_Float16`` type on SSE2 and above following X86 psABI. 252(`D107082 <https://reviews.llvm.org/D107082>`_) 253 254The change may affect the current use of ``half`` includes (but is not limited 255to): 256 257* Frontends generating ``half`` type in function passing and/or returning 258 arguments. 259* Downstream runtimes providing any ``half`` conversion builtins assuming the 260 old ABI. 261* Projects built with LLVM 15.0 but using early versions of compiler-rt. 262 263When you find failures with ``half`` type, check the calling conversion of the 264code and switch it to the new ABI. 265 266Changes to the OCaml bindings 267----------------------------- 268 269 270Changes to the C API 271-------------------- 272 273* Add ``LLVMGetCastOpcode`` function to aid users of ``LLVMBuildCast`` in 274 resolving the best cast operation given a source value and destination type. 275 This function is a direct wrapper of ``CastInst::getCastOpcode``. 276 277* Add ``LLVMGetAggregateElement`` function as a wrapper for 278 ``Constant::getAggregateElement``, which can be used to fetch an element of a 279 constant struct, array or vector, independently of the underlying 280 representation. The ``LLVMGetElementAsConstant`` function is deprecated in 281 favor of the new function, which works on all constant aggregates, rather than 282 only instances of ``ConstantDataSequential``. 283 284* The following functions for creating constant expressions have been removed, 285 because the underlying constant expressions are no longer supported. Instead, 286 an instruction should be created using the ``LLVMBuildXYZ`` APIs, which will 287 constant fold the operands if possible and create an instruction otherwise: 288 289 * ``LLVMConstExtractValue`` 290 * ``LLVMConstInsertValue`` 291 * ``LLVMConstUDiv`` 292 * ``LLVMConstExactUDiv`` 293 * ``LLVMConstSDiv`` 294 * ``LLVMConstExactSDiv`` 295 * ``LLVMConstURem`` 296 * ``LLVMConstSRem`` 297 * ``LLVMConstFAdd`` 298 * ``LLVMConstFSub`` 299 * ``LLVMConstFMul`` 300 * ``LLVMConstFDiv`` 301 * ``LLVMConstFRem`` 302 303* Add ``LLVMDeleteInstruction`` function which allows deleting instructions that 304 are not inserted into a basic block. 305 306* As part of the opaque pointer migration, the following APIs are deprecated and 307 will be removed in the next release: 308 309 * ``LLVMBuildLoad`` -> ``LLVMBuildLoad2`` 310 * ``LLVMBuildCall`` -> ``LLVMBuildCall2`` 311 * ``LLVMBuildInvoke`` -> ``LLVMBuildInvoke2`` 312 * ``LLVMBuildGEP`` -> ``LLVMBuildGEP2`` 313 * ``LLVMBuildInBoundsGEP`` -> ``LLVMBuildInBoundsGEP2`` 314 * ``LLVMBuildStructGEP`` -> ``LLVMBuildStructGEP2`` 315 * ``LLVMBuildPtrDiff`` -> ``LLVMBuildPtrDiff2`` 316 * ``LLVMConstGEP`` -> ``LLVMConstGEP2`` 317 * ``LLVMConstInBoundsGEP`` -> ``LLVMConstInBoundsGEP2`` 318 * ``LLVMAddAlias`` -> ``LLVMAddAlias2`` 319 320* Refactor compression namespaces across the project, making way for a possible 321 introduction of alternatives to zlib compression in the llvm toolchain. 322 Changes are as follows: 323 324 * Relocate the ``llvm::zlib`` namespace to ``llvm::compression::zlib``. 325 * Remove crc32 from zlib compression namespace, people should use the ``llvm::crc32`` instead. 326 327Changes to the Go bindings 328-------------------------- 329 330 331Changes to the FastISel infrastructure 332-------------------------------------- 333 334* ... 335 336Changes to the DAG infrastructure 337--------------------------------- 338 339 340Changes to the Metadata Info 341--------------------------------- 342 343* Add Module Flags Metadata ``stack-protector-guard-symbol`` which specify a 344 symbol for addressing the stack-protector guard. 345 346Changes to the Debug Info 347--------------------------------- 348 349During this release ... 350 351Changes to the LLVM tools 352--------------------------------- 353 354* (Experimental) :doc:`llvm-symbolizer <CommandGuide/llvm-symbolizer>` now has ``--filter-markup`` to 355 filter :doc:`Symbolizer Markup </SymbolizerMarkupFormat>` into human-readable 356 form. 357* :doc:`llvm-objcopy <CommandGuide/llvm-objcopy>` has removed support for the legacy ``zlib-gnu`` format. 358* :doc:`llvm-objcopy <CommandGuide/llvm-objcopy>` now allows ``--set-section-flags src=... --rename-section src=tst``. 359 ``--add-section=.foo1=... --rename-section=.foo1=.foo2`` now adds ``.foo1`` instead of ``.foo2``. 360* New features supported on AIX for ``llvm-ar``: 361 362 * AIX big-format archive write operation (`D123949 <https://reviews.llvm.org/D123949>`_) 363 364 * A new object mode option, ``-X`` , to specify the type of object file ``llvm-ar`` should operate upon (`D127864 <https://reviews.llvm.org/D127864>`_) 365 366 * Read global symbols of AIX big archive (`D124865 <https://reviews.llvm.org/D124865>`_) 367 368* New options supported for ``llvm-nm``: 369 370 * ``-X``, to specify the type of object file that ``llvm-nm`` should examine (`D118193 <https://reviews.llvm.org/D118193>`_) 371 372 * ``--export-symbols``, to create a list of symbols to export (`D112735 <https://reviews.llvm.org/D112735>`_) 373 374* The LLVM gold plugin now ignores bitcode from the ``.llvmbc`` section of ELF 375 files when doing LTO. https://github.com/llvm/llvm-project/issues/47216 376* llvm-objcopy now supports 32 bit XCOFF. 377* llvm-objdump: improved assembly printing for XCOFF. 378* llc now parses code-model attribute from input file. 379 380Changes to LLDB 381--------------------------------- 382 383* The "memory region" command now has a "--all" option to list all 384 memory regions (including unmapped ranges). This is the equivalent 385 of using address 0 then repeating the command until all regions 386 have been listed. 387* Added "--show-tags" option to the "memory find" command. This is off by default. 388 When enabled, if the target value is found in tagged memory, the tags for that 389 memory will be shown inline with the memory contents. 390* Various memory related parts of LLDB have been updated to handle 391 non-address bits (such as AArch64 pointer signatures): 392 393 * "memory read", "memory write" and "memory find" can now be used with 394 addresses with non-address bits. 395 * All the read and write memory methods on SBProccess and SBTarget can 396 be used with addreses with non-address bits. 397 * When printing a pointer expression, LLDB can now dereference the result 398 even if it has non-address bits. 399 * The memory cache now ignores non-address bits when looking up memory 400 locations. This prevents us reading locations multiple times, or not 401 writing out new values if the addresses have different non-address bits. 402 403* LLDB now supports reading memory tags from AArch64 Linux core files. 404 405* LLDB now supports the gnu debuglink section for reading debug information 406 from a separate file on Windows 407 408* LLDB now allows selecting the C++ ABI to use on Windows (between Itanium, 409 used for MingW, and MSVC) via the ``plugin.object-file.pe-coff.abi`` setting. 410 In Windows builds of LLDB, this defaults to the style used for LLVM's default 411 target. 412 413Changes to Sanitizers 414--------------------- 415 416 417Other Changes 418------------- 419* The code for the `LLVM Visual Studio integration 420 <https://marketplace.visualstudio.com/items?itemName=LLVMExtensions.llvm-toolchain>`_ 421 has been removed. This had been obsolete and abandoned since Visual Studio 422 started including an integration by default in 2019. 423 424* Added the unwinder, personality, and helper functions for exception handling 425 on AIX. (`D100132 <https://reviews.llvm.org/D100132>`_) 426 (`D100504 <https://reviews.llvm.org/D100504>`_) 427 428* PGO on AIX: A new implementation that requires linker support 429 (__start_SECTION/__stop_SECTION symbols) available on AIX 7.2 TL5 SP4 and 430 AIX 7.3 TL0 SP2. 431 432External Open Source Projects Using LLVM 15 433=========================================== 434 435* A project... 436 437Additional Information 438====================== 439 440A wide variety of additional information is available on the `LLVM web page 441<https://llvm.org/>`_, in particular in the `documentation 442<https://llvm.org/docs/>`_ section. The web page also contains versions of the 443API documentation which is up-to-date with the Git version of the source 444code. You can access versions of these documents specific to this release by 445going into the ``llvm/docs/`` directory in the LLVM tree. 446 447If you have any questions or comments about LLVM, please feel free to contact 448us via the `mailing lists <https://llvm.org/docs/#mailing-lists>`_. 449