This document contains the release notes for the LLVM Compiler Infrastructure, release 2.9. Here we describe the status of LLVM, including major improvements from the previous release and significant known problems. 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 LLVM Developer's Mailing List is a good place to send them.

Note that if you are reading this file from a Subversion 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.

The LLVM 2.9 distribution currently consists of code from the core LLVM repository (which roughly includes the LLVM optimizers, code generators and supporting tools), the Clang repository and the llvm-gcc repository. In addition to this code, the LLVM Project includes other sub-projects that are in development. Here we include updates on these subprojects.

Clang is an LLVM front end for the C, C++, and Objective-C languages. Clang aims to provide a better user experience through expressive diagnostics, a high level of conformance to language standards, fast compilation, and low memory use. Like LLVM, Clang provides a modular, library-based architecture that makes it suitable for creating or integrating with other development tools. Clang is considered a production-quality compiler for C, Objective-C, C++ and Objective-C++ on x86 (32- and 64-bit), and for darwin/arm targets.

In the LLVM 2.9 time-frame, the Clang team has made many improvements in C, C++ and Objective-C support. C++ support is now generally rock solid, has been exercised on a broad variety of code, and has several new C++'0x features implemented (such as rvalue references and variadic templates). LLVM 2.9 has also brought in a large range of bug fixes and minor features (e.g. __label__ support), and is much more compatible with the Linux Kernel.

If Clang rejects your code but another compiler accepts it, please take a look at the language compatibility guide to make sure this is not intentional or a known issue.

DragonEgg is a gcc plugin that replaces GCC's optimizers and code generators with LLVM's. Currently it requires a patched version of gcc-4.5. The plugin can target the x86-32 and x86-64 processor families and has been used successfully on the Darwin, FreeBSD and Linux platforms. The Ada, C, C++ and Fortran languages work well. The plugin is capable of compiling plenty of Obj-C, Obj-C++ and Java but it is not known whether the compiled code actually works or not!

The 2.9 release has the following notable changes:

• The plugin is much more stable when compiling Fortran.
• Inline assembly where an asm output is tied to an input of a different size is now supported in many more cases.
• Basic support for the __float128 type was added. It is now possible to generate LLVM IR from programs using __float128 but code generation does not work yet.
• Compiling Java programs no longer systematically crashes the plugin.

The new LLVM compiler-rt project is a simple library that provides an implementation of the low-level target-specific hooks required by code generation and other runtime components. For example, when compiling for a 32-bit target, converting a double to a 64-bit unsigned integer is compiled into a runtime call to the "__fixunsdfdi" function. The compiler-rt library provides highly optimized implementations of this and other low-level routines (some are 3x faster than the equivalent libgcc routines).

In the LLVM 2.9 timeframe, compiler_rt has had several minor changes for better ARM support, and a fairly major license change. All of the code in the compiler-rt project is now dual licensed under MIT and UIUC license, which allows you to use compiler-rt in applications without the binary copyright reproduction clause. If you prefer the LLVM/UIUC license, you are free to continue using it under that license as well.

LLDB is a brand new member of the LLVM umbrella of projects. LLDB is a next generation, high-performance debugger. It is built as a set of reusable components which highly leverage existing libraries in the larger LLVM Project, such as the Clang expression parser, the LLVM disassembler and the LLVM JIT.

LLDB is has advanced by leaps and bounds in the 2.9 timeframe. It is dramatically more stable and useful, and includes both a new tutorial and a side-by-side comparison with GDB.

libc++ is another new member of the LLVM family. It is an implementation of the C++ standard library, written from the ground up to specifically target the forthcoming C++'0X standard and focus on delivering great performance.

In the LLVM 2.9 timeframe, libc++ has had numerous bugs fixed, and is now being co-developed with Clang's C++'0x mode.

Like compiler_rt, libc++ is now dual licensed under the MIT and UIUC license, allowing it to be used more permissively.

LLBrowse is an interactive viewer for LLVM modules. It can load any LLVM module and displays its contents as an expandable tree view, facilitating an easy way to inspect types, functions, global variables, or metadata nodes. It is fully cross-platform, being based on the popular wxWidgets GUI toolkit.

The VMKit project is an implementation of a Java Virtual Machine (Java VM or JVM) that uses LLVM for static and just-in-time compilation. As of LLVM 2.9, VMKit now supports generational garbage collectors. The garbage collectors are provided by the MMTk framework, and VMKit can be configured to use one of the numerous implemented collectors of MMTk.

An exciting aspect of LLVM is that it is used as an enabling technology for a lot of other language and tools projects. This section lists some of the projects that have already been updated to work with LLVM 2.9.

Crack aims to provide the ease of development of a scripting language with the performance of a compiled language. The language derives concepts from C++, Java and Python, incorporating object-oriented programming, operator overloading and strong typing.

TCE is a toolset for designing application-specific processors (ASP) based on the Transport triggered architecture (TTA). The toolset provides a complete co-design flow from C/C++ programs down to synthesizable VHDL and parallel program binaries. Processor customization points include the register files, function units, supported operations, and the interconnection network.

TCE uses Clang and LLVM for C/C++ language support, target independent optimizations and also for parts of code generation. It generates new LLVM-based code generators "on the fly" for the designed TTA processors and loads them in to the compiler backend as runtime libraries to avoid per-target recompilation of larger parts of the compiler chain.

PinaVM is an open source, SystemC front-end. Unlike many other front-ends, PinaVM actually executes the elaboration of the program analyzed using LLVM's JIT infrastructure. It later enriches the bitcode with SystemC-specific information.

Pure is an algebraic/functional programming language based on term rewriting. Programs are collections of equations which are used to evaluate expressions in a symbolic fashion. The interpreter uses LLVM as a backend to JIT-compile Pure programs to fast native code. Pure offers dynamic typing, eager and lazy evaluation, lexical closures, a hygienic macro system (also based on term rewriting), built-in list and matrix support (including list and matrix comprehensions) and an easy-to-use interface to C and other programming languages (including the ability to load LLVM bitcode modules, and inline C, C++, Fortran and Faust code in Pure programs if the corresponding LLVM-enabled compilers are installed).

Pure version 0.47 has been tested and is known to work with LLVM 2.9 (and continues to work with older LLVM releases >= 2.5).

IcedTea provides a harness to build OpenJDK using only free software build tools and to provide replacements for the not-yet free parts of OpenJDK. One of the extensions that IcedTea provides is a new JIT compiler named Shark which uses LLVM to provide native code generation without introducing processor-dependent code.

OpenJDK 7 b112, IcedTea6 1.9 and IcedTea7 1.13 and later have been tested and are known to work with LLVM 2.9 (and continue to work with older LLVM releases >= 2.6 as well).

GHC is an open source, state-of-the-art programming suite for Haskell, a standard lazy functional programming language. It includes an optimizing static compiler generating good code for a variety of platforms, together with an interactive system for convenient, quick development.

In addition to the existing C and native code generators, GHC 7.0 now supports an LLVM code generator. GHC supports LLVM 2.7 and later.

Polly is a project that aims to provide advanced memory access optimizations to better take advantage of SIMD units, cache hierarchies, multiple cores or even vector accelerators for LLVM. Built around an abstract mathematical description based on Z-polyhedra, it provides the infrastructure to develop advanced optimizations in LLVM and to connect complex external optimizers. In its first year of existence Polly already provides an exact value-based dependency analysis as well as basic SIMD and OpenMP code generation support. Furthermore, Polly can use PoCC(Pluto) an advanced optimizer for data-locality and parallelism.

Rubinius is an environment for running Ruby code which strives to write as much of the implementation in Ruby as possible. Combined with a bytecode interpreting VM, it uses LLVM to optimize and compile ruby code down to machine code. Techniques such as type feedback, method inlining, and deoptimization are all used to remove dynamism from ruby execution and increase performance.

FAUST is a compiled language for real-time audio signal processing. The name FAUST stands for Functional AUdio STream. Its programming model combines two approaches: functional programming and block diagram composition. In addition with the C, C++, JAVA output formats, the Faust compiler can now generate LLVM bitcode, and works with LLVM 2.7-2.9.

This release includes a huge number of bug fixes, performance tweaks and minor improvements. Some of the major improvements and new features are listed in this section.

LLVM 2.9 includes several major new capabilities:

• Type Based Alias Analysis (TBAA) is now implemented and turned on by default in Clang. This allows substantially better load/store optimization in some cases. TBAA can be disabled by passing -fno-strict-aliasing.
• This release has seen a continued focus on quality of debug information. LLVM now generates much higher fidelity debug information, particularly when debugging optimized code.
• Inline assembly now supports multiple alternative constraints.
• A new backend for the NVIDIA PTX virtual ISA (used to target its GPUs) is under rapid development. It is not generally useful in 2.9, but is making rapid progress.

LLVM IR has several new features for better support of new targets and that expose new optimization opportunities:

• The udiv, ashr, lshr, and shl instructions now have support exact and nuw/nsw bits to indicate that they don't overflow or shift out bits. This is useful for optimization of pointer differences and other cases.
• LLVM IR now supports the unnamed_addr attribute to indicate that constant global variables with identical initializers can be merged. This fixed an issue where LLVM would incorrectly merge two globals which were supposed to have distinct addresses.
• The new hotpatch attribute has been added to allow runtime patching of functions.

In addition to a large array of minor performance tweaks and bug fixes, this release includes a few major enhancements and additions to the optimizers:

• Link Time Optimization (LTO) has been improved to use MC for parsing inline assembly and now can build large programs like Firefox 4 on both Mac OS X and Linux.
• The new -loop-idiom pass recognizes memset/memcpy loops (and memset_pattern on darwin), turning them into library calls, which are typically better optimized than inline code. If you are building a libc and notice that your memcpy and memset functions are compiled into infinite recursion, please build with -ffreestanding or -fno-builtin to disable this pass.
• A new -early-cse pass does a fast pass over functions to fold constants, simplify expressions, perform simple dead store elimination, and perform common subexpression elimination. It does a good job at catching some of the trivial redundancies that exist in unoptimized code, making later passes more effective.
• A new -loop-instsimplify pass is used to clean up loop bodies in the loop optimizer.
• The new TargetLibraryInfo interface allows mid-level optimizations to know whether the current target's runtime library has certain functions. For example, the optimizer can now transform integer-only printf calls to call iprintf, allowing reduced code size for embedded C libraries (e.g. newlib).
• LLVM has a new RegionPass infrastructure for region-based optimizations.
• Several optimizer passes have been substantially sped up: GVN is much faster on functions with deep dominator trees and lots of basic blocks. The dominator tree and dominance frontier passes are much faster to compute, and preserved by more passes (so they are computed less often). The -scalar-repl pass is also much faster and doesn't use DominanceFrontier.
• The Dead Store Elimination pass is more aggressive optimizing stores of different types: e.g. a large store following a small one to the same address. The MemCpyOptimizer pass handles several new forms of memcpy elimination.
• LLVM now optimizes various idioms for overflow detection into check of the flag register on various CPUs. For example, we now compile:

     unsigned long t = a+b;
     if (t < a) ...
    

  into:

     addq %rdi, %rbx
     jno  LBB0_2
    

The LLVM Machine Code (aka MC) subsystem was created to solve a number of problems in the realm of assembly, disassembly, object file format handling, and a number of other related areas that CPU instruction-set level tools work in.

• ELF MC support has matured enough for the integrated assembler to be turned on by default in Clang on X86-32 and X86-64 ELF systems.
• MC supports and CodeGen uses the .file and .loc directives for producing line number debug info. This produces more compact line tables and easier to read .s files.
• MC supports the .cfi_* directives for producing DWARF frame information, but it is still not used by CodeGen by default.
• The MC assembler now generates much better diagnostics for common errors, is much faster at matching instructions, is much more bug-compatible with the GAS assembler, and is now generally useful for a broad range of X86 assembly.
• We now have some basic internals documentation for MC.
• .td files can now specify assembler aliases directly with the MnemonicAlias and InstAlias tblgen classes.
• LLVM now has an experimental format-independent object file manipulation library (lib/Object). It supports both PE/COFF and ELF. The llvm-nm tool has been extended to work with native object files, and the new llvm-objdump tool supports disassembly of object files (but no relocations are displayed yet).
• Win32 PE-COFF support in the MC assembler has made a lot of progress in the 2.9 timeframe, but is still not generally useful.

For more information, please see the Intro to the LLVM MC Project Blog Post.

We have put a significant amount of work into the code generator infrastructure, which allows us to implement more aggressive algorithms and make it run faster:

• The pre-register-allocation (preRA) instruction scheduler models register pressure much more accurately in some cases. This allows the adoption of more aggressive scheduling heuristics without causing spills to be generated.
• LiveDebugVariables is a new pass that keeps track of debugging information for user variables that are promoted to registers in optimized builds.
• The scheduler now models operand latency and pipeline forwarding.
• A major register allocator infrastructure rewrite is underway. It is not on by default for 2.9 and you are not advised to use it, but it has made substantial progress in the 2.9 timeframe:
  • A new -regalloc=basic "basic" register allocator can be used as a simple fallback when debugging. It uses the new infrastructure.
  • New infrastructure is in place for live range splitting. "SplitKit" can break a live interval into smaller pieces while preserving SSA form, and SpillPlacement can help find the best split points. This is a work in progress so the API is changing quickly.
  • The inline spiller has learned to clean up after live range splitting. It can hoist spills out of loops, and it can eliminate redundant spills.
  • Rematerialization works with live range splitting.
  • The new "greedy" register allocator using live range splitting. This will be the default register allocator in the next LLVM release, but it is not turned on by default in 2.9.

New features and major changes in the X86 target include:

• LLVM 2.9 includes a complete reimplementation of the MMX instruction set. The reimplementation uses a new LLVM IR x86_mmx type to ensure that MMX operations are only generated from source that uses MMX builtin operations. With this, random types like <2 x i32> are not turned into MMX operations (which can be catastrophic without proper "emms" insertion). Because the X86 code generator always generates reliable code, the -disable-mmx flag is now removed.
• X86 support for FS/GS relative loads and stores using address space 256/257 works reliably now.
• LLVM 2.9 generates much better code in several cases by using adc/sbb to avoid generation of conditional move instructions for conditional increment and other idioms.
• The X86 backend has adopted a new preRA scheduling mode, "list-ilp", to shorten the height of instruction schedules without inducing register spills.
• The MC assembler supports 3dNow! and 3DNowA instructions.
• Several bugs have been fixed for Windows x64 code generator.

New features of the ARM target include:

• The ARM backend now has a fast instruction selector, which dramatically improves -O0 compile times.
• The ARM backend has new tuning for Cortex-A8 and Cortex-A9 CPUs.
• The __builtin_prefetch builtin (and llvm.prefetch intrinsic) is compiled into prefetch instructions instead of being discarded.
• The ARM backend preRA scheduler now models machine resources at cycle granularity. This allows the scheduler to both accurately model instruction latency and avoid overcommitting functional units.
• Countless ARM microoptimizations have landed in LLVM 2.9.

• MicroBlaze: major updates for aggressive delay slot filler, MC-based assembly printing, assembly instruction parsing, ELF .o file emission, and MC instruction disassembler have landed.
• SPARC: Many improvements, including using the Y registers for multiplications and addition of a simple delay slot filler.
• PowerPC: The backend has been largely MC'ized and is ready to support directly writing out mach-o object files. No one seems interested in finishing this final step though.
• Mips: Improved o32 ABI support, including better varags handling. More instructions supported in codegen: madd, msub, rotr, rotrv and clo. It also now supports lowering block addresses.

If you're already an LLVM user or developer with out-of-tree changes based on LLVM 2.8, this section lists some "gotchas" that you may run into upgrading from the previous release.

• This is the last release to support the llvm-gcc frontend.
• LLVM has a new naming convention standard, though the codebase hasn't fully adopted it yet.
• The new DIBuilder class provides a simpler interface for front ends to encode debug info in LLVM IR, and has replaced DIFactory.
• LLVM IR and other tools always work on normalized target triples (which have been run through Triple::normalize).
• The target triple x86_64--mingw64 is obsoleted. Use x86_64--mingw32 instead.
• The PointerTracking pass has been removed from mainline, and moved to The ClamAV project (its only client).
• The LoopIndexSplit, LiveValues, SimplifyHalfPowrLibCalls, GEPSplitter, and PartialSpecialization passes were removed. They were unmaintained, buggy, or deemed to be a bad idea.

In addition, many APIs have changed in this release. Some of the major LLVM API changes are:

• include/llvm/System merged into include/llvm/Support.
• The llvm::APInt API was significantly cleaned up.
• In the code generator, MVT::Flag was renamed to MVT::Glue to more accurately describe its behavior.
• The system_error header from C++0x was added, and is now pervasively used to capture and handle i/o and other errors in LLVM.
• The old sys::Path API has been deprecated in favor of the new PathV2 API, which is more efficient and flexible.

This section contains significant known problems with the LLVM system, listed by component. If you run into a problem, please check the LLVM bug database and submit a bug if there isn't already one.

The following components of this LLVM release are either untested, known to be broken or unreliable, or are in early development. These components should not be relied on, and bugs should not be filed against them, but they may be useful to some people. In particular, if you would like to work on one of these components, please contact us on the LLVMdev list.

• The Alpha, Blackfin, CellSPU, MicroBlaze, MSP430, MIPS, PTX, SystemZ and XCore backends are experimental.
• llc "-filetype=obj" is experimental on all targets other than darwin and ELF X86 systems.

• The X86 backend does not yet support all inline assembly that uses the X86 floating point stack. It supports the 'f' and 't' constraints, but not 'u'.
• The X86-64 backend does not yet support the LLVM IR instruction va_arg. Currently, front-ends support variadic argument constructs on X86-64 by lowering them manually.
• Windows x64 (aka Win64) code generator has a few issues.
  • llvm-gcc cannot build the mingw-w64 runtime currently due to lack of support for the 'u' inline assembly constraint and for X87 floating point inline assembly.
  • On mingw-w64, you will see unresolved symbol __chkstk due to Bug 8919. It is fixed in r128206.
  • Miss-aligned MOVDQA might crash your program. It is due to Bug 9483, lack of handling aligned internal globals.

• The Linux PPC32/ABI support needs testing for the interpreter and static compilation, and lacks support for debug information.

• Thumb mode works only on ARMv6 or higher processors. On sub-ARMv6 processors, thumb programs can crash or produce wrong results (PR1388).
• Compilation for ARM Linux OABI (old ABI) is supported but not fully tested.

• The SPARC backend only supports the 32-bit SPARC ABI (-m32); it does not support the 64-bit SPARC ABI (-m64).

• 64-bit MIPS targets are not supported yet.

• On 21164s, some rare FP arithmetic sequences which may trap do not have the appropriate nops inserted to ensure restartability.

The C backend has numerous problems and is not being actively maintained. Depending on it for anything serious is not advised.

• The C backend has only basic support for inline assembly code.
• The C backend violates the ABI of common C++ programs, preventing intermixing between C++ compiled by the CBE and C++ code compiled with llc or native compilers.
• The C backend does not support all exception handling constructs.
• The C backend does not support arbitrary precision integers.

LLVM 2.9 will be the last release of llvm-gcc.

llvm-gcc is generally very stable for the C family of languages. The only major language feature of GCC not supported by llvm-gcc is the __builtin_apply family of builtins. However, some extensions are only supported on some targets. For example, trampolines are only supported on some targets (these are used when you take the address of a nested function).

Fortran support generally works, but there are still several unresolved bugs in Bugzilla. Please see the tools/gfortran component for details. Note that llvm-gcc is missing major Fortran performance work in the frontend and library that went into GCC after 4.2. If you are interested in Fortran, we recommend that you consider using dragonegg instead.

The llvm-gcc 4.2 Ada compiler has basic functionality, but is no longer being actively maintained. If you are interested in Ada, we recommend that you consider using dragonegg instead.

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 Subversion version of the source code. You can access versions of these documents specific to this release by going into the "llvm/doc/" directory in the LLVM tree.

If you have any questions or comments about LLVM, please feel free to contact us via the mailing lists.