3 <style> .red {color:red} </style>
14 Written by the `LLVM Team <http://llvm.org/>`_
16 :red:`These are in-progress notes for the upcoming LLVM 3.2 release. You may
17 prefer the` `LLVM 3.1 Release Notes <http://llvm.org/releases/3.1/docs
18 /ReleaseNotes.html>`_.
23 This document contains the release notes for the LLVM Compiler Infrastructure,
24 release 3.2. Here we describe the status of LLVM, including major improvements
25 from the previous release, improvements in various subprojects of LLVM, and
26 some of the current users of the code. All LLVM releases may be downloaded
27 from the `LLVM releases web site <http://llvm.org/releases/>`_.
29 For more information about LLVM, including information about the latest
30 release, please check out the `main LLVM web site <http://llvm.org/>`_. If you
31 have questions or comments, the `LLVM Developer's Mailing List
32 <http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev>`_ is a good place to send
35 Note that if you are reading this file from a Subversion checkout or the main
36 LLVM web page, this document applies to the *next* release, not the current
37 one. To see the release notes for a specific release, please see the `releases
38 page <http://llvm.org/releases/>`_.
40 Sub-project Status Update
41 =========================
43 The LLVM 3.2 distribution currently consists of code from the core LLVM
44 repository, which roughly includes the LLVM optimizers, code generators and
45 supporting tools, and the Clang repository. In addition to this code, the LLVM
46 Project includes other sub-projects that are in development. Here we include
47 updates on these subprojects.
49 Clang: C/C++/Objective-C Frontend Toolkit
50 -----------------------------------------
52 `Clang <http://clang.llvm.org/>`_ is an LLVM front end for the C, C++, and
53 Objective-C languages. Clang aims to provide a better user experience through
54 expressive diagnostics, a high level of conformance to language standards, fast
55 compilation, and low memory use. Like LLVM, Clang provides a modular,
56 library-based architecture that makes it suitable for creating or integrating
57 with other development tools. Clang is considered a production-quality
58 compiler for C, Objective-C, C++ and Objective-C++ on x86 (32- and 64-bit), and
59 for Darwin/ARM targets.
61 In the LLVM 3.2 time-frame, the Clang team has made many improvements.
64 #. More powerful warnings, especially `-Wuninitialized`
65 #. Template type diffing in diagnostic messages
66 #. Higher quality and more efficient debug info generation
68 For more details about the changes to Clang since the 3.1 release, see the
69 `Clang release notes. <http://clang.llvm.org/docs/ReleaseNotes.html>`_
71 If Clang rejects your code but another compiler accepts it, please take a look
72 at the `language compatibility <http://clang.llvm.org/compatibility.html>`_
73 guide to make sure this is not intentional or a known issue.
75 DragonEgg: GCC front-ends, LLVM back-end
76 ----------------------------------------
78 `DragonEgg <http://dragonegg.llvm.org/>`_ is a `gcc plugin
79 <http://gcc.gnu.org/wiki/plugins>`_ that replaces GCC's optimizers and code
80 generators with LLVM's. It works with gcc-4.5 and gcc-4.6 (and partially with
81 gcc-4.7), can target the x86-32/x86-64 and ARM processor families, and has been
82 successfully used on the Darwin, FreeBSD, KFreeBSD, Linux and OpenBSD
83 platforms. It fully supports Ada, C, C++ and Fortran. It has partial support
84 for Go, Java, Obj-C and Obj-C++.
86 The 3.2 release has the following notable changes:
88 #. Able to load LLVM plugins such as Polly.
89 #. Supports thread-local storage models.
90 #. Passes knowledge of variable lifetimes to the LLVM optimizers.
91 #. No longer requires GCC to be built with LTO support.
93 compiler-rt: Compiler Runtime Library
94 -------------------------------------
96 The new LLVM `compiler-rt project <http://compiler-rt.llvm.org/>`_ is a simple
97 library that provides an implementation of the low-level target-specific hooks
98 required by code generation and other runtime components. For example, when
99 compiling for a 32-bit target, converting a double to a 64-bit unsigned integer
100 is compiled into a runtime call to the ``__fixunsdfdi`` function. The
101 ``compiler-rt`` library provides highly optimized implementations of this and
102 other low-level routines (some are 3x faster than the equivalent libgcc
105 The 3.2 release has the following notable changes:
109 LLDB: Low Level Debugger
110 ------------------------
112 `LLDB <http://lldb.llvm.org>`_ is a ground-up implementation of a command line
113 debugger, as well as a debugger API that can be used from other applications.
114 LLDB makes use of the Clang parser to provide high-fidelity expression parsing
115 (particularly for C++) and uses the LLVM JIT for target support.
117 The 3.2 release has the following notable changes:
121 libc++: C++ Standard Library
122 ----------------------------
124 Like compiler_rt, libc++ is now :ref:`dual licensed
125 <copyright-license-patents>` under the MIT and UIUC license, allowing it to be
126 used more permissively.
128 Within the LLVM 3.2 time-frame there were the following highlights:
135 The `VMKit project <http://vmkit.llvm.org/>`_ is an implementation of a Java
136 Virtual Machine (Java VM or JVM) that uses LLVM for static and just-in-time
139 The 3.2 release has the following notable changes:
143 Polly: Polyhedral Optimizer
144 ---------------------------
146 `Polly <http://polly.llvm.org/>`_ is an *experimental* optimizer for data
147 locality and parallelism. It provides high-level loop optimizations and
148 automatic parallelisation.
150 Within the LLVM 3.2 time-frame there were the following highlights:
152 #. isl, the integer set library used by Polly, was relicensed to the MIT license
153 #. isl based code generation
154 #. MIT licensed replacement for CLooG (LGPLv2)
155 #. Fine grained option handling (separation of core and border computations,
156 control overhead vs. code size)
157 #. Support for FORTRAN and dragonegg
158 #. OpenMP code generation fixes
160 External Open Source Projects Using LLVM 3.2
161 ============================================
163 An exciting aspect of LLVM is that it is used as an enabling technology for a
164 lot of other language and tools projects. This section lists some of the
165 projects that have already been updated to work with LLVM 3.2.
170 `Crack <http://code.google.com/p/crack-language/>`_ aims to provide the ease of
171 development of a scripting language with the performance of a compiled
172 language. The language derives concepts from C++, Java and Python,
173 incorporating object-oriented programming, operator overloading and strong
179 `FAUST <http://faust.grame.fr/>`_ is a compiled language for real-time audio
180 signal processing. The name FAUST stands for Functional AUdio STream. Its
181 programming model combines two approaches: functional programming and block
182 diagram composition. In addition with the C, C++, Java, JavaScript output
183 formats, the Faust compiler can generate LLVM bitcode, and works with LLVM
186 Glasgow Haskell Compiler (GHC)
187 ------------------------------
189 `GHC <http://www.haskell.org/ghc/>`_ is an open source compiler and programming
190 suite for Haskell, a lazy functional programming language. It includes an
191 optimizing static compiler generating good code for a variety of platforms,
192 together with an interactive system for convenient, quick development.
194 GHC 7.0 and onwards include an LLVM code generator, supporting LLVM 2.8 and
200 `Julia <https://github.com/JuliaLang/julia>`_ is a high-level, high-performance
201 dynamic language for technical computing. It provides a sophisticated
202 compiler, distributed parallel execution, numerical accuracy, and an extensive
203 mathematical function library. The compiler uses type inference to generate
204 fast code without any type declarations, and uses LLVM's optimization passes
205 and JIT compiler. The `Julia Language <http://julialang.org/>`_ is designed
206 around multiple dispatch, giving programs a large degree of flexibility. It is
207 ready for use on many kinds of problems.
212 `LLVM D Compiler <https://github.com/ldc-developers/ldc>`_ (LDC) is a compiler
213 for the D programming Language. It is based on the DMD frontend and uses LLVM
216 Open Shading Language
217 ---------------------
219 `Open Shading Language (OSL)
220 <https://github.com/imageworks/OpenShadingLanguage/>`_ is a small but rich
221 language for programmable shading in advanced global illumination renderers and
222 other applications, ideal for describing materials, lights, displacement, and
223 pattern generation. It uses LLVM to JIT complex shader networks to x86 code at
226 OSL was developed by Sony Pictures Imageworks for use in its in-house renderer
227 used for feature film animation and visual effects, and is distributed as open
228 source software with the "New BSD" license.
230 Portable OpenCL (pocl)
231 ----------------------
233 In addition to producing an easily portable open source OpenCL implementation,
234 another major goal of `pocl <http://pocl.sourceforge.net/>`_ is improving
235 performance portability of OpenCL programs with compiler optimizations,
236 reducing the need for target-dependent manual optimizations. An important part
237 of pocl is a set of LLVM passes used to statically parallelize multiple
238 work-items with the kernel compiler, even in the presence of work-group
239 barriers. This enables static parallelization of the fine-grained static
240 concurrency in the work groups in multiple ways (SIMD, VLIW, superscalar, ...).
245 `Pure <http://pure-lang.googlecode.com/>`_ is an algebraic/functional
246 programming language based on term rewriting. Programs are collections of
247 equations which are used to evaluate expressions in a symbolic fashion. The
248 interpreter uses LLVM as a backend to JIT-compile Pure programs to fast native
249 code. Pure offers dynamic typing, eager and lazy evaluation, lexical closures,
250 a hygienic macro system (also based on term rewriting), built-in list and
251 matrix support (including list and matrix comprehensions) and an easy-to-use
252 interface to C and other programming languages (including the ability to load
253 LLVM bitcode modules, and inline C, C++, Fortran and Faust code in Pure
254 programs if the corresponding LLVM-enabled compilers are installed).
256 Pure version 0.54 has been tested and is known to work with LLVM 3.1 (and
257 continues to work with older LLVM releases >= 2.5).
259 TTA-based Co-design Environment (TCE)
260 -------------------------------------
262 `TCE <http://tce.cs.tut.fi/>`_ is a toolset for designing application-specific
263 processors (ASP) based on the Transport triggered architecture (TTA). The
264 toolset provides a complete co-design flow from C/C++ programs down to
265 synthesizable VHDL/Verilog and parallel program binaries. Processor
266 customization points include the register files, function units, supported
267 operations, and the interconnection network.
269 TCE uses Clang and LLVM for C/C++ language support, target independent
270 optimizations and also for parts of code generation. It generates new
271 LLVM-based code generators "on the fly" for the designed TTA processors and
272 loads them in to the compiler backend as runtime libraries to avoid per-target
273 recompilation of larger parts of the compiler chain.
275 Installation Instructions
276 =========================
278 See :doc:`GettingStarted`.
280 What's New in LLVM 3.2?
281 =======================
283 This release includes a huge number of bug fixes, performance tweaks and minor
284 improvements. Some of the major improvements and new features are listed in
292 Features that need text if they're finished for 3.2:
296 loop dependence analysis
297 CorrelatedValuePropagation
298 Integrated assembler on by default for arm/thumb?
301 Analysis/RegionInfo.h + Dom Frontiers
302 SparseBitVector: used in LiveVar.
303 llvm/lib/Archive - replace with lib object?
306 LLVM 3.2 includes several major changes and big features:
308 #. New NVPTX back-end (replacing existing PTX back-end) based on NVIDIA sources
311 LLVM IR and Core Improvements
312 -----------------------------
314 LLVM IR has several new features for better support of new targets and that
315 expose new optimization opportunities:
317 #. Thread local variables may have a specified TLS model. See the :ref:`Language
318 Reference Manual <globalvars>`.
321 Optimizer Improvements
322 ----------------------
324 In addition to many minor performance tweaks and bug fixes, this release
325 includes a few major enhancements and additions to the optimizers:
327 Loop Vectorizer - We've added a loop vectorizer and we are now able to
328 vectorize small loops. The loop vectorizer is disabled by default and can be
329 enabled using the ``-mllvm -vectorize-loops`` flag. The SIMD vector width can
330 be specified using the flag ``-mllvm -force-vector-width=4``. The default
331 value is ``0`` which means auto-select.
333 We can now vectorize this function:
337 unsigned sum_arrays(int *A, int *B, int start, int end) {
339 for (int i = start; i < end; ++i)
340 sum += A[i] + B[i] + i;
344 We vectorize under the following loops:
346 #. The inner most loops must have a single basic block.
347 #. The number of iterations are known before the loop starts to execute.
348 #. The loop counter needs to be incremented by one.
349 #. The loop trip count **can** be a variable.
350 #. Loops do **not** need to start at zero.
351 #. The induction variable can be used inside the loop.
352 #. Loop reductions are supported.
353 #. Arrays with affine access pattern do **not** need to be marked as
354 '``noalias``' and are checked at runtime.
357 SROA - We've re-written SROA to be significantly more powerful and generate
358 code which is much more friendly to the rest of the optimization pipeline.
359 Previously this pass had scaling problems that required it to only operate on
360 relatively small aggregates, and at times it would mistakenly replace a large
361 aggregate with a single very large integer in order to make it a scalar SSA
362 value. The result was a large number of i1024 and i2048 values representing any
363 small stack buffer. These in turn slowed down many subsequent optimization
366 The new SROA pass uses a different algorithm that allows it to only promote to
367 scalars the pieces of the aggregate actively in use. Because of this it doesn't
368 require any thresholds. It also always deduces the scalar values from the uses
369 of the aggregate rather than the specific LLVM type of the aggregate. These
370 features combine to both optimize more code with the pass but to improve the
371 compile time of many functions dramatically.
373 #. Branch weight metadata is preseved through more of the optimizer.
376 MC Level Improvements
377 ---------------------
379 The LLVM Machine Code (aka MC) subsystem was created to solve a number of
380 problems in the realm of assembly, disassembly, object file format handling,
381 and a number of other related areas that CPU instruction-set level tools work
382 in. For more information, please see the `Intro to the LLVM MC Project Blog
383 Post <http://blog.llvm.org/2010/04/intro-to-llvm-mc-project.html>`_.
389 Target Independent Code Generator Improvements
390 ----------------------------------------------
392 We have put a significant amount of work into the code generator
393 infrastructure, which allows us to implement more aggressive algorithms and
398 Stack Coloring - We have implemented a new optimization pass to merge stack
399 objects which are used in disjoin areas of the code. This optimization reduces
400 the required stack space significantly, in cases where it is clear to the
401 optimizer that the stack slot is not shared. We use the lifetime markers to
402 tell the codegen that a certain alloca is used within a region.
404 We now merge consecutive loads and stores.
406 X86-32 and X86-64 Target Improvements
407 -------------------------------------
409 New features and major changes in the X86 target include:
415 ARM Target Improvements
416 -----------------------
418 New features of the ARM target include:
422 .. _armintegratedassembler:
424 MIPS Target Improvements
425 ------------------------
427 New features and major changes in the MIPS target include:
431 PowerPC Target Improvements
432 ---------------------------
434 Many fixes and changes across LLVM (and Clang) for better compliance with the
435 64-bit PowerPC ELF Application Binary Interface, interoperability with GCC, and
436 overall 64-bit PowerPC support. Some highlights include:
438 #. MCJIT support added.
439 #. PPC64 relocation support and (small code model) TOC handling added.
440 #. Parameter passing and return value fixes (alignment issues, padding, varargs
441 support, proper register usage, odd-sized structure support, float support,
442 extension of return values for i32 return values).
443 #. Fixes in spill and reload code for vector registers.
444 #. C++ exception handling enabled.
445 #. Changes to remediate double-rounding compatibility issues with respect to
447 #. Refactoring to disentangle ``ppc64-elf-linux`` ABI from Darwin ppc64 ABI
449 #. Assorted new test cases and test case fixes (endian and word size issues).
450 #. Fixes for big-endian codegen bugs, instruction encodings, and instruction
452 #. Implemented ``-integrated-as`` support.
453 #. Additional support for Altivec compare operations.
454 #. IBM long double support.
456 There have also been code generation improvements for both 32- and 64-bit code.
457 Instruction scheduling support for the Freescale e500mc and e5500 cores has
460 PTX/NVPTX Target Improvements
461 -----------------------------
463 The PTX back-end has been replaced by the NVPTX back-end, which is based on the
464 LLVM back-end used by NVIDIA in their CUDA (nvcc) and OpenCL compiler. Some
467 #. Compatibility with PTX 3.1 and SM 3.5.
468 #. Support for NVVM intrinsics as defined in the NVIDIA Compiler SDK.
469 #. Full compatibility with old PTX back-end, with much greater coverage of LLVM
472 Please submit any back-end bugs to the LLVM Bugzilla site.
474 Other Target Specific Improvements
475 ----------------------------------
479 Major Changes and Removed Features
480 ----------------------------------
482 If you're already an LLVM user or developer with out-of-tree changes based on
483 LLVM 3.2, this section lists some "gotchas" that you may run into upgrading
484 from the previous release.
486 #. The CellSPU port has been removed. It can still be found in older versions.
492 In addition, many APIs have changed in this release. Some of the major LLVM
495 We've added a new interface for allowing IR-level passes to access
496 target-specific information. A new IR-level pass, called
497 ``TargetTransformInfo`` provides a number of low-level interfaces. LSR and
498 LowerInvoke already use the new interface.
500 The ``TargetData`` structure has been renamed to ``DataLayout`` and moved to
501 ``VMCore`` to remove a dependency on ``Target``.
508 In addition, some tools have changed in this release. Some of the changes are:
515 Officially supported Python bindings have been added! Feature support is far
516 from complete. The current bindings support interfaces to:
523 LLVM is generally a production quality compiler, and is used by a broad range
524 of applications and shipping in many products. That said, not every subsystem
525 is as mature as the aggregate, particularly the more obscure1 targets. If you
526 run into a problem, please check the `LLVM bug database
527 <http://llvm.org/bugs/>`_ and submit a bug if there isn't already one or ask on
528 the `LLVMdev list <http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev>`_.
530 Known problem areas include:
532 #. The MSP430 and XCore backends are experimental.
534 #. The integrated assembler, disassembler, and JIT is not supported by several
535 targets. If an integrated assembler is not supported, then a system
536 assembler is required. For more details, see the
537 :ref:`target-feature-matrix`.
539 Additional Information
540 ======================
542 A wide variety of additional information is available on the `LLVM web page
543 <http://llvm.org/>`_, in particular in the `documentation
544 <http://llvm.org/docs/>`_ section. The web page also contains versions of the
545 API documentation which is up-to-date with the Subversion version of the source
546 code. You can access versions of these documents specific to this release by
547 going into the ``llvm/docs/`` directory in the LLVM tree.
549 If you have any questions or comments about LLVM, please feel free to contact
550 us via the `mailing lists <http://llvm.org/docs/#maillist>`_.