```{warning} These are in-progress notes for the upcoming LLVM {{env.config.release}} release. Release notes for previous releases can be found on [the Download Page](https://releases.llvm.org/download.html). ```
This document contains the release notes for the LLVM Compiler Infrastructure, release {{env.config.release}}. Here we describe the status of LLVM, including major improvements from the previous release, improvements in various subprojects of LLVM, and some of the current users of the code. All LLVM releases may be downloaded from the LLVM releases web site.
For more information about LLVM, including information about the latest release, please check out the main LLVM web site. If you have questions or comments, the Discourse forums is a good place to ask them.
Note that if you are reading this file from a Git checkout or the main LLVM 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.
Added a new IRNormalizer pass which aims to transform LLVM modules into a normal form by reordering and renaming instructions while preserving the same semantics. The normalizer makes it easier to spot semantic differences when diffing two modules which have undergone different passes.
The SPIR-V backend is now an official LLVM target, providing OpenCL and SYCL conformance and establishing a foundation for broader applicability to other APIs, including Vulkan, GLSL, and HLSL. This backend aims to offer a unified approach for diverse compute and graphics workloads, providing a robust alternative to the Khronos SPIR-V LLVM Translator.
...
Types are no longer allowed to be recursive.
The x86_mmx IR type has been removed. It will be translated to the standard vector type <1 x i64> in bitcode upgrade.
Renamed llvm.experimental.stepvector intrinsic to llvm.stepvector.
Added usub_cond and usub_sat operations to atomicrmw.
Introduced noalias.addrspace metadata.
Remove the following intrinsics which can be replaced with a bitcast:
llvm.nvvm.bitcast.f2illvm.nvvm.bitcast.i2fllvm.nvvm.bitcast.d2llllvm.nvvm.bitcast.ll2dRemove the following intrinsics which can be replaced with a funnel-shift:
llvm.nvvm.rotate.b32llvm.nvvm.rotate.right.b64llvm.nvvm.rotate.b64Remove the following intrinsics which can be replaced with an addrspacecast:
llvm.nvvm.ptr.gen.to.globalllvm.nvvm.ptr.gen.to.sharedllvm.nvvm.ptr.gen.to.constantllvm.nvvm.ptr.gen.to.localllvm.nvvm.ptr.global.to.genllvm.nvvm.ptr.shared.to.genllvm.nvvm.ptr.constant.to.genllvm.nvvm.ptr.local.to.genRemove the following intrinsics which can be relaced with a load from addrspace(1) with an !invariant.load metadata
llvm.nvvm.ldg.global.illvm.nvvm.ldg.global.fllvm.nvvm.ldg.global.pOperand bundle values can now be metadata strings.
Fast math flags are now permitted on fptrunc and fpext.
BasicBlock::iterators instead. The pointer-flavoured methods will be removed in a future release. This work is part of the RemoveDIs project, the documentation for which contains instructions for updating call-sites using the deprecated methods.LLVM_ENABLE_PROJECTS. Users should instead use LLVM_ENABLE_RUNTIMES, either through the runtimes or the bootstrapping build.LLVM_ENABLE_PROJECTS. Users should instead use LLVM_ENABLE_RUNTIMES, either through the runtimes or the bootstrapping build..balign N, 0, .p2align N, 0, .align N, 0 in code sections will now fill the required alignment space with a sequence of 0x0 bytes (the requested fill value) rather than NOPs.
Assembler/disassembler support has been added for Armv9.6-A (2024) architecture extensions.
Added support for the FUJITSU-MONAKA CPU.
Updated feature dependency in Armv9.6 for FEAT_FAMINMAX, FEAT_LUT and FEAT_FP8, now they depend only on FEAT_NEON.
Initial support for gfx950
Improved llvm.memcpy, llvm.memmove and llvm.memset lowering
Fixed expansion of 64-bit flat address space atomicrmw and cmpxchg operations which may access private memory. noalias.addrspace metadat may be used to avoid the expansion if the target address is known to not be on the stack.
Fix compile failures when emitting unreachable functions.
Removed llvm.amdgcn.flat.atomic.fadd and llvm.amdgcn.global.atomic.fadd intrinsics. Users should use the {ref}atomicrmw <i_atomicrmw> instruction with fadd and addrspace(0) or addrspace(1) instead.
.balign N, 0, .p2align N, 0, .align N, 0 in code sections will now fill the required alignment space with a sequence of 0x0 bytes (the requested fill value) rather than NOPs.
The default behavior for frame pointers in leaf functions has been updated. When the -fno-omit-frame-pointer option is specified, FPKeepKindStr is set to -mframe-pointer=all, meaning the frame pointer (FP) is now retained in leaf functions by default. To eliminate the frame pointer in leaf functions, you must explicitly use the -momit-leaf-frame-pointer option.
When using the MOVT or MOVW instructions, the Assembler will now check to ensure that any addend that is used is within a 16-bit signed value range. If the addend falls outside of this range, the LLVM backend will emit an error like so Relocation Not In Range.
Incorrect GOT usage for non-dso_local function calls in large code model is fixed.
A gprof support issue is fixed.
A SDAG hang issue caused by ISD::CONCAT_VECTORS is fixed.
A compiler crash issue when converting half to i32 is fixed.
Almost all of la64v1.1 instructions can now be generated. The full list is frecipe.s, frecipe.d, frsqrte.s, frsqrte.d, vfrecipe.s, vfrecipe.d, vfrsqrte.s, vfrsqrte.d, xvfrecipe.s, xvfrecipe.d, xvfrsqrte.s, xvfrsqrte.d, sc.q, amcas.b, amcas.h, amcas.w, amcas.d, amcas_db.b, amcas_db.h, amcas_db.w, amcas_db.d, amswap.b, amswap.h, amswap_db.b, amswap_db.h, amadd.b, amadd.h, amadd_db.b, amadd_db.h. Optionally generate instructions dbar 0x700, div.w, div.wu, mod.w and mod.wu when related target features are enabled. llacq.w, screl.w, llacq.d and screl.d cannot be generated yet.
An llc option called -loongarch-annotate-tablejump is added to annotate table jump instruction in the .discard.tablejump_annotate section. A typical user of these annotations is the objtool in Linux kernel.
The default cpu in MCSubtargetInfo is changed from la464 to generic-la64. In addition, the lsx feature is added to generic-la64.
CFI instructions now allow register names and aliases, previously only numbers were allowed.
RuntimeDyld now supports LoongArch, which means that programs relying on MCJIT can now work.
.balign N, 0, .p2align N, 0, .align N, 0 in code sections will now fill the required alignment space with a sequence of 0x0 bytes (the requested fill value) rather than NOPs.
%ld_pcrel_20, %gd_pcrel_20 and %desc_pcrel_20 operand modifiers are supported by assembler.
A machine function pass called LoongArch Merge Base Offset is added to merge the offset of address calculation into the offset field of instructions in a global address lowering sequence.
The LoopDataPrefetch pass can now work on LoongArch, but it is disabled by default due to the bad effect on Fortran benchmarks.
Enable alias analysis by default.
Avoid indirect branch jumps using the $ra register.
Other optimizations.
ppc64 LLC default cpu is updated from ppc to ppc64.AIX Specific:
generic to pwr7..balign N, 0, .p2align N, 0, .align N, 0 in code sections will now fill the required alignment space with a sequence of 0x0 bytes (the requested fill value) rather than NOPs.-mcpu=syntacore-scr4/5-rv32/64-mcpu=sifive-p470 was added.-mcpu=rp2350-hazard3 (32-bit only).Zvbc32e and Zvkgs extensions are now supported experimentally.Smctr, Ssctr and Svvptc extensions.-mcpu=syntacore-scr7 was added.-mcpu=tt-ascalon-d8 was added.-mcpu=mips-p8700 was added.-mcpu=sifive-p550 was added.Zacas extension is no longer marked as experimental.Smmpm, Smnpm, Ssnpm, Supm, and Sspm pointer masking extensions are no longer marked as experimental.Sha extension is now supported..insn <length>, <raw encoding> can be used to assemble 48- and 64-bit instructions from raw integer values..insn [<length>,] <raw encoding> now accepts absolute expressions for both expressions, so that they can be computed from constants and absolute symbols.cr constraint meaning an RVC-encoding compatible GPR (x8-x15)cf constraint meaning an RVC-encoding compatible FPR (f8-f15)R constraint meaning an even-odd GPR pair (prints as the even register, but both registers in the pair are considered live).cR constraint meaning an RVC-encoding compatible even-odd GPR Pair (prints as an even register between x8 and x14, but both registers in the pair are considered live).N modifer meaning print the register encoding (0-31) rather than the name.f and cf inline assembly constraints, when using F-/D-/H-in-X extensions, will use the relevant GPR rather than FPR. This makes inline assembly portable between e.g. F and Zfinx code.Sdext and Sdtrig extensions.__builtin_setjump and __builtin_longjmp.The default target CPU, “generic”, now enables the -mnontrapping-fptoint and -mbulk-memory flags, which correspond to the Bulk Memory Operations and Non-trapping float-to-int Conversions language features, which are widely implemented in engines.
A new Lime1 target CPU is added, -mcpu=lime1. This CPU follows the definition of the Lime1 CPU here, and enables -mmultivalue, -mmutable-globals, -mcall-indirect-overlong, -msign-ext, -mbulk-memory-opt, -mnontrapping-fptoint, and -mextended-const.
Support for the new standardized Exception Handling proposal is added. The legacy Exception Handling proposal is still supported, and turned on by the newly added -wasm-use-legacy-eh option. Given that major web browsers still default to the legacy EH proposal, this option is turned on by default for the moment.
.balign N, 0x90, .p2align N, 0x90, and .align N, 0x90 in code sections now fill the required alignment space with repeating 0x90 bytes, rather than using optimised NOP filling. Optimised NOP filling fills the space with NOP instructions of various widths, not just those that use the 0x90 byte encoding. To use optimised NOP filling in a code section, leave off the “fillval” argument, i.e. .balign N, .p2align N or .align N respectively.
Due to the removal of the x86_mmx IR type, functions with x86_mmx arguments or return values will use a different, incompatible, calling convention ABI. Such functions are not generally seen in the wild (Clang never generates them!), so this is not expected to result in real-world compatibility problems.
Support ISA of AVX10.2-256 and AVX10.2-512.
Supported instructions of MOVRS AND AVX10.2
Supported ISA of SM4(EVEX).
Supported ISA of MSR_IMM.
Supported -mcpu=diamondrapids
Supported emitting relocation types for x86-64 target:
R_X86_64_CODE_4_GOTPCRELXR_X86_64_CODE_4_GOTTPOFFR_X86_64_CODE_4_GOTPC32_TLSDESCR_X86_64_CODE_6_GOTTPOFFThe following symbols are deleted due to the removal of the x86_mmx IR type:
LLVMX86_MMXTypeKindLLVMX86MMXTypeInContextLLVMX86MMXTypeThe following functions are added to further support non-null-terminated strings:
LLVMGetNamedFunctionWithLength
LLVMGetNamedGlobalWithLength
The following functions are added to access the LLVMContextRef associated with LLVMValueRef and LLVMBuilderRef objects:
LLVMGetValueContextLLVMGetBuilderContextThe new pass manager can now be invoked with a custom alias analysis pipeline, using the LLVMPassBuilderOptionsSetAAPipeline function.
It is now also possible to run the new pass manager on a single function, by calling LLVMRunPassesOnFunction instead of LLVMRunPasses.
Support for creating instructions with custom synchronization scopes has been added:
LLVMGetSyncScopeID to map a synchronization scope name to an ID.LLVMBuildFenceSyncScope, LLVMBuildAtomicRMWSyncScope and LLVMBuildAtomicCmpXchgSyncScope versions of the existing builder functions with an additional synchronization scope ID parameter.LLVMGetAtomicSyncScopeID and LLVMSetAtomicSyncScopeID to get and set the synchronization scope of any atomic instruction.LLVMIsAtomic to check if an instruction is atomic, for use with the above functions. Because of backwards compatibility, LLVMIsAtomicSingleThread and LLVMSetAtomicSingleThread continue to work with any instruction type.The LLVMSetPersonalityFn and LLVMSetInitializer APIs now support clearing the personality function and initializer respectively by passing a null pointer.
The following functions are added to allow iterating over debug records attached to instructions:
LLVMGetFirstDbgRecordLLVMGetLastDbgRecordLLVMGetNextDbgRecordLLVMGetPreviousDbgRecordAdded LLVMAtomicRMWBinOpUSubCond and LLVMAtomicRMWBinOpUSubSat to LLVMAtomicRMWBinOp enum for AtomicRMW instructions.
target attribute, and command line options).fmv-features. The metadata string value consists of a comma-separated list of Function Multi Versioning feature names as defined in the Arm C Language Extensions (ACLE).llvm-objcopy now supports the following options for Mach-O: --globalize-symbol, --globalize-symbols, --keep-global-symbol, --keep-global-symbols, --localize-symbol, --localize-symbols, --skip-symbol, --skip-symbols.
llvm-objcopy now prints the correct file path in the error message when the output file specified by --dump-section cannot be opened.
llvm-cxxfilt now supports demangling call expressions encoded using cp instead of cl.
llvm-objdump now supports printing the file header, load section header and auxiliary header for XCOFF object files under the --private-headers option.
It is now recommended that LLDB be built with Python >= 3.8, but no changes have been made to the supported Python versions. The next release, LLDB 21, will require Python >= 3.8.
LLDB now supports inline diagnostics for the expression evaluator and command line parser.
Old:
(lldb) p a+b
error: <user expression 0>:1:1: use of undeclared identifier 'a'
1 | a+b
| ^
error: <user expression 0>:1:3: use of undeclared identifier 'b'
1 | a+b
| ^
New:
(lldb) p a+b
˄ ˄
│ ╰─ error: use of undeclared identifier 'b'
╰─ error: use of undeclared identifier 'a'
Program stdout/stderr redirection will now open the file with O_TRUNC flag, make sure to truncate the file if path already exists.
settings set target.output-path/target.error-path <path/to/file>A new setting target.launch-working-dir can be used to set a persistent cwd that is used by default by process launch and run.
LLDB now parses shared libraries in parallel, resulting in an average 2x speedup when attaching (only available on Darwin platforms) and launching (available on all platforms).
It is now possible to implement lldb commands in Python that use lldb's native command-line parser. In particular, that allows per-option/argument completion, with all the basic completers automatically supported and auto-generated help. The command template file in lldb/examples/python/cmdtemplate.py has been updated to show how to use this.
Breakpoints on “inlined call sites” are now supported. Previous to this fix, breakpoints on source lines that only contained inlined call sites would be moved to the next source line, causing you to miss the inlined executions.
On the command line, LLDB now limits tab completions to your terminal width to avoid wrapping.
Old:
Available completions:
_regexp-attach -- Attach to process by ID or name.
_regexp-break -- Set a breakpoint using one of several shorthand
formats.
_regexp-bt -- Show backtrace of the current thread's call sta
ck. Any numeric argument displays at most that many frames. The argument 'al
l' displays all threads. Use 'settings set frame-format' to customize the pr
inting of individual frames and 'settings set thread-format' to customize th
e thread header. Frame recognizers may filter thelist. Use 'thread backtrace
-u (--unfiltered)' to see them all.
_regexp-display -- Evaluate an expression at every stop (see 'help
target stop-hook'.)
New:
Available completions:
_regexp-attach -- Attach to process by ID or name.
_regexp-break -- Set a breakpoint using one of several shorth...
_regexp-bt -- Show backtrace of the current thread's call ...
_regexp-display -- Evaluate an expression at every stop (see 'h...
DWARF indexing speed (for binaries not using the debug_names index) increased by 30-60%.
The frame diagnose now works on ELF-based systems. After a crash, LLDB will try to determine the likely cause of the signal, matching Darwin behavior. This feature requires using a new lldb-server version and (like Darwin) only works on x86 binaries.
* thread #1, name = 'a.out', stop reason = signal SIGSEGV: address not mapped to object (fault address=0x4)
frame #0: 0x00005555555551aa a.out`GetSum(f=0x0000555555558018) at main.c:21:37
18 }
19
20 int GetSum(struct Foo *f) {
-> 21 return SumTwoIntegers(f->a, f->b->d ? 0 : 1);
22 }
23
24 int main() {
Likely cause: f->b->d accessed 0x4
Minidumps generated by LLDB now support:
New Core File API. This gives greater control on the data captured into the core file, relative to the existing process save-core styles.
When opening ELF core files, LLDB will print additional information about the signal that killed the process and the disassembly view will display actual (relocated) targets of the jump instructions instead of raw offsets encoded in the instruction. This matches existing behavior for live processes.
Old:
* thread #1: tid = 329384, 0x0000000000401262, name = 'a.out', stop reason = signal SIGSEGV 0x7f1e3193e0a7 <+23>: ja 0xfe100 ; <+112>
New:
* thread #1: tid = 329384, 0x0000000000401262, name = 'a.out', stop reason = SIGSEGV: address not mapped to object (fault address=0x0) 0x7f1e3193e0a7 <+23>: ja 0x7f1e3193e100 ; <+112>
lldb-server now listens to a single port for gdbserver connections and provides that port to the connection handler processes. This means that only 2 ports need to be opened in the firewall (one for the lldb-server platform, one for gdbserver connections). In addition, due to this work, lldb-server now works on Windows in the server mode.
LLDB can now read the fpmr register from AArch64 Linux processes and core files.
Support was added for debugging AArch64 Linux programs that use the Guarded Control Stack extension (GCS). This includes live processes and core files.
LLDB now supports execution of user expressions for non-trivial cases for LoongArch and RISC-V targets, like function calls, when some code needs to be executed on the target.
LLDB now supports optionally enabled/disabled register sets (particularly floating point registers) for RISC-V 64. This happens for targets like RV64IMAC or RV64IMACV, that have no floating point registers. The change is applied to native debugging and core-file usage.
LLDB now supports core-file for LoongArch.
LLDB now supports hardware breakpoint and watchpoint for LoongArch.
LLDB now supports vector registers for LoongArch when debugging a live process.
Incorrect floating-point register DWARF numbers for LoongArch were fixed.
Support was added for handling the GDB Remote Protocol x packet in the format introduced by GDB 16.2. LLDB currently uses a different format for x and LLDB is now able to handle both formats. At some point in the future support for LLDB's format of x will be removed.
atexit-registered functions are not called when a DSO is dlclose'd, a mechanism is added that implements this missing functionality for calls to atexit in the profile runtime. This is currently only enabled on AIX.A wide variety of additional information is available on the LLVM web page, in particular in the documentation section. The web page also contains versions of the API documentation which is 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 llvm/docs/ directory in the LLVM tree.
If you have any questions or comments about LLVM, please feel free to contact us via the Discourse forums.