Written by the LLVM Team
+
Introduction
@@ -32,7 +33,7 @@
- Evan added support for simple exit value substitution to LSR.
-
-
-
-
@@ -300,9 +349,9 @@ optimizer, speeding it up and making it more aggressive. For example:
@@ -372,6 +420,8 @@ components, please contact us on the alignment
problems on operating systems that don't require 16-byte stack alignment
(including most non-darwin OS's like linux).
+This document contains the release notes for the LLVM compiler -infrastructure, release 2.1. Here we describe the status of LLVM, including +infrastructure, release 2.2. Here we describe the status of LLVM, including major improvements from the previous release and any known problems. All LLVM releases may be downloaded from the LLVM releases web site.
@@ -58,31 +59,197 @@ current one. To see the release notes for a specific releases, please see the
-
+
+
+
+
+
+
+This is the twelfth public release of the LLVM Compiler Infrastructure. -It includes many features and refinements from LLVM 2.0.
+This is the thirteenth public release of the LLVM Compiler Infrastructure. +It includes many features and refinements from LLVM 2.1.
+ +
+
+
This is the last LLVM release to support llvm-gcc 4.0, llvm-upgrade, and +llvmc in its current form. llvm-gcc 4.0 has been replaced with llvm-gcc 4.2. +llvm-upgrade is useful for upgrading llvm 1.9 files to llvm 2.x syntax, but you +can always use an old release to do this. llvmc is currently mostly useless in +llvm 2.2, and will be redesigned or removed in llvm 2.3.
-
+
+
+
+
+LLVM 2.1 brings two new beta C front-ends. First, a new version of llvm-gcc -based on GCC 4.2, innovatively called "llvm-gcc-4.2". This promises to bring -FORTRAN and Ada support to LLVM as well as features like atomic builtins and -OpenMP. None of these actually work yet, but don't let that stop you checking -it out!
+LLVM 2.2 fully supports both the llvm-gcc 4.0 and llvm-gcc 4.2 front-ends (in +LLVM 2.1, llvm-gcc 4.2 was beta). Since LLVM 2.1, the llvm-gcc 4.2 front-end +has made leaps and bounds and is now at least as good as 4.0 in virtually every +area, and is better in several areas (for example, exception handling +correctness, support for Ada and Fortran, better ABI compatibility, etc). We +strongly recommend that you +migrate from llvm-gcc 4.0 to llvm-gcc 4.2 in this release cycle because +LLVM 2.2 is the last release that will support llvm-gcc 4.0: LLVM 2.3 +will only support the llvm-gcc 4.2 front-end.
-Second, LLVM now includes its own native C and Objective-C front-end (C++ is -in progress, but is not very far along) code named "clang". This front-end has a number of great -features, primarily aimed at source-level analysis and speeding up compile-time. -At this point though, the LLVM Code Generator component is still very early in -development, so it's mostly useful for people looking to build source-level -analysis tools or source-to-source translators.
+The clang project is an effort to build +a set of new 'llvm native' front-end technologies for the LLVM optimizer +and code generator. Currently, its C and Objective-C support is maturing +nicely, and it has advanced source-to-source analysis and transformation +capabilities. If you are interested in building source-level tools for C and +Objective-C (and eventually C++), you should take a look. However, note that +clang is not an official part of the LLVM 2.2 release. If you are interested in +this project, please see its web site.
+ +
+
+
+
+
+
+
+LLVM 2.2 includes several major new capabilities:
+ +-
+
- A research team led by Scott Michel in the Computer Systems Research +Department at The Aerospace Corporation contributed the CellSPU backend, which +generates code for the vector coprocessors on the Sony/Toshiba/IBM Cell BE +processor. llvm-gcc 4.2 supports CellSPU as a 'configure' target and progress +is being made so that libgcc.a compiles cleanly. Notable pieces still in +development include full 64-bit integer and full double precision floating +point support. + +
- Anton and Duncan significantly improved llvm-gcc 4.2 support for the GCC Ada +(GNAT) and Fortran (gfortran) front-ends. These front-ends should still be considered +experimental however: see the list of known problems. +The release binaries do not contain either front-end: they need to be built from +source (the Ada front-end only builds on x86-32 linux). + +
- Dale contributed full support for long double on x86/x86-64 (where it is 80 +bits) and on Darwin PPC/PPC64 (where it is 128 bits). In previous LLVM +releases, llvm-gcc silently mapped long double to double. + +
- Gordon Henriksen rewrote most of the Accurate Garbage Collection code in the code generator, making the +generated code more efficient and adding support for the OCaml garbage collector +metadata format. + +
- Christopher Lamb contributed support for multiple address spaces in LLVM +IR. This is useful for supporting targets that have 'near' vs 'far' pointers, +'RAM' vs 'ROM' pointers, or that have non-local memory that can be accessed with +special instructions. + +
- LLVM now includes a new set of detailed tutorials, which explain how to implement a +language with LLVM and shows how to use several important APIs. + +
+
+
+
+
+
+New features include: +
+ +-
+
- Gordon contributed support for C and OCaml Bindings for the basic LLVM IR +construction routines as well as several other auxiliary APIs. + +
- Anton added readnone/readonly attributes for modeling function side effects. +Duncan hooked up GCC's pure/const attributes to them and enhanced mod/ref +analysis to use them. + +
- Devang added LLVMFoldingBuilder, a version of LLVMBuilder that implicitly +simplifies the code as it is constructed. + +
- Ted Kremenek added a framework for generic object serialization to bitcode +files. This support is only used by clang right now for ASTs but is extensible +and could be used for serializing arbitrary other data into bitcode files. + +
- Duncan improved TargetData to distinguish between the size/alignment of a +type in a register, in memory according to the platform ABI, and in memory when +we have a choice. + +
- Reid moved parameter attributes off of FunctionType and onto functions +and calls. This makes it much easier to add attributes to a function in a +transformation pass. + +
- Dan Gohman added support for vector sin, cos, and pow intrinsics. + +
+
+
@@ -93,78 +260,24 @@ analysis tools or source-to-source translators.
We put a significant amount of work into the code generator infrastructure, +which allows us to implement more aggressive algorithms and make it run +faster:
+ +-
+
+
- Owen refactored the existing LLVM dominator and loop information code to +allow it work on the machine code representation. He contributed support for +dominator and loop information on machine code and merged the code for forward +and backward dominator computation. + +
- Dan added support for emitting debug information with .file and .loc +directives on platforms that support it, instead of emitting large tables in the .s +file. + +
- Evan extended the DAG scheduler to model physical register dependencies +explicitly and have the BURR scheduler pick a correct schedule based on the +dependencies. This reduces our use of the 'flag' operand hack. + +
- Evan added initial support for register coalescing of subregister +references. + +
- Rafael Espindola implemented initial support for a new 'byval' attribute, +which allows more efficient by-value argument passing in the LLVM IR. Evan +finished support for it and enabled it in the X86 (32- and 64-bit) and C +backends. + +
- The LLVM TargetInstrInfo class can now answer queries about the mod/ref and +side-effect behavior of MachineInstr's. This information is inferred +automatically by TableGen from .td files for all instructions with +patterns. + +
- Evan implemented simple live interval splitting on basic block boundaries. +This allows the register allocator to be more successful at keeping values in +registers in some parts of a value's live range, even if they need to be spilled +in some other block. + +
- The new MachineRegisterInfo.h class provides support for efficiently +iterating over all defs/uses of a register, and this information is +automatically kept up-to-date. This support is similar to the use_iterator in +the LLVM IR level. + +
- The MachineInstr, MachineOperand and TargetInstrDesc classes are simpler, +more consistent, and better documented. +
-- Dan Gohman contributed several enhancements to Loop Strength Reduction (LSR)
+to make it more aggressive with SSE intrinsics and when induction variables are
+used by non-memory instructions.
-
+Some of the most noticable feature improvements this release have been in the -optimizer, speeding it up and making it more aggressive. For example:
+In addition to a huge array of bug fixes and minor performance tweaks, the +LLVM 2.2 optimizers support a few major enhancements:
-
-
- Owen Anderson wrote the new MemoryDependenceAnalysis pass, which provides - a lazy, caching layer on top of AliasAnalysis. He then used it to rewrite - DeadStoreElimination which resulted in significantly better compile time in - common cases, -
- Owen implemented the new GVN pass, which is also based on - MemoryDependenceAnalysis. This pass replaces GCSE/LoadVN in the standard - set of passes, providing more aggressive optimization at a some-what - improved compile-time cost. -
- Owen implemented GVN-PRE, a partial redundancy elimination algorithm that - shares some details with the new GVN pass. It is still in need of compile - time tuning, and is not turned on by default. -
- Devang merged ETForest and DomTree into a single easier to use data - structure. This makes it more obvious which datastructure to choose - (because there is only one) and makes the compiler more memory and time - efficient (less stuff to keep up-to-date). -
- Nick Lewycky improved loop trip count analysis to handle many more common - cases. +
- Daniel Berlin and Curtis Dunham rewrote Andersen's alias analysis to be +several orders of magnitude faster, and implemented Offline Variable +Substitution and Lazy Cycle Detection. Note that Andersen's is not enabled in +llvm-gcc by default, but can be accessed through 'opt'. -
-
-
One of the main focuses of this release was performance tuning and bug - fixing. In addition to these, several new major changes occurred:
- --
-
-
- Dale finished up the Tail Merging optimization in the code generator, and - enabled it by default. This produces smaller code that is also faster in - some cases. - -
- Christopher Lamb implemented support for virtual register sub-registers, - which can be used to better model many forms of subregisters. As an example - use, he modified the X86 backend to use this to model truncates and - extends more accurately (leading to better code). - -
- Dan Gohman changed the way we represent vectors before legalization, - significantly simplifying the SelectionDAG representation for these and - making the code generator faster for vector code. - -
- Evan contributed a new target independent if-converter. While it is - target independent, so far only the ARM backend uses it. - -
- Evan rewrote the way the register allocator handles rematerialization, - allowing it to be much more effective on two-address targets like X86, - and taught it to fold loads away when possible (also a big win on X86). - -
- Dan Gohman contributed support for better alignment and volatility handling - in the code generator, and significantly enhanced alignment analysis for SSE - load/store instructions. With his changes, an insufficiently-aligned SSE - load instruction turns into movups, for example. - -
- Duraid Madina contributed a new "bigblock" register allocator, and Roman - Levenstein contributed several big improvements. BigBlock is optimized for - code that uses very large basic blocks. It is slightly slower than the - "local" allocator, but produces much better code. - -
- David Greene refactored the register allocator to split coalescing out from - allocation, making coalescers pluggable. +
- Evan enhanced LSR to support induction variable reuse when the induction +variables have different widths.
-
-
-
-
-
-
-New features include: +
New target-specific features include:
-
-
- Bruno Cardoso Lopes contributed initial MIPS support. It is sufficient to - run many small programs, but is still incomplete and is not yet - fully performant. - -
- Bill Wendling added SSSE3 support to the X86 backend. - -
- Nicholas Geoffray contributed improved linux/ppc ABI and JIT support. - -
- Dale Johannesen rewrote handling of 32-bit float values in the X86 backend - when using the floating point stack, fixing several nasty bugs. - -
- Dan contributed rematerialization support for the X86 backend, in addition - to several X86-specific micro optimizations. -
-
-
-
-
-
-
-New features include: -
- --
-
- Duncan and Anton made significant progress chasing down a number of problems - with C++ Zero-Cost exception handling in llvm-gcc 4.0 and 4.2. It is now at - the point where it "just works" on linux/X86-32 and has partial support on - other targets. - -
- Devang and Duncan fixed a huge number of bugs relating to bitfields, pragma - pack, and variable sized fields in structures. - -
- Tanya implemented support for __attribute__((noinline)) in - llvm-gcc, and added support for generic variable annotations which are - propagated into the LLVM IR, e.g. - "int X __attribute__((annotate("myproperty")));". - -
- Sheng Zhou and Christopher Lamb implemented alias analysis support for -"restrict" pointer arguments to functions. - -
- Duncan contributed support for trampolines (taking the address of a nested - function). Currently this is only supported on the X86-32 target. - -
- Lauro Ramos Venancio contributed support to encode alignment info in - load and store instructions, the foundation for other alignment-related - work. -
-
@@ -276,13 +320,18 @@ optimizer, speeding it up and making it more aggressive. For example:
New features include: -
- --
-
- Neil Booth contributed a new "APFloat" class, which ensures that floating - point representation and constant folding is not dependent on the host - architecture that builds the application. This support is the foundation - for "long double" support that will be wrapped up in LLVM 2.2. - -
- Based on the APFloat class, Dale redesigned the internals of the ConstantFP - class and has been working on extending the core and optimizer components to - support various target-specific 'long double's. We expect this work to be - completed in LLVM 2.2. - -
- LLVM now provides an LLVMBuilder class, which makes it significantly easier - to create LLVM IR instructions. - -
- Reid contributed support for intrinsics that take arbitrary integer typed - arguments. Dan Gohman and Chandler extended it to support arbitrary - floating point arguments and vectors. +
- Evan contributed support to the X86 backend to model the mod/ref behavior +of the EFLAGS register explicitly in all instructions. This gives more freedom +to the scheduler, and is a more explicit way to model the instructions. +
- Dale contributed support for exception handling on Darwin/PPC and he and +Anton got x86-64 working. +
- Evan turned on if-conversion by default for ARM, allowing LLVM to take +advantage of its predication features. +
- Bruno added PIC support to the MIPS backend, fixed many bugs and improved +support for architecture variants. +
- Arnold Schwaighofer added initial support for X86 tail calls. +
- Evan contributed several enhancements to Darwin/x86 debug information. +
- Duncan added x86-64 support for trampolines (pointers to nested functions).
-
-
- Sterling Stein contributed a new BrainF frontend, located in llvm/examples. - This shows a some of the more modern APIs for building a front-end, and - demonstrates JIT compiler support. +
- Gordon expanded and updated the LLVM Analysis and +Transformation Passes reference to include descriptions for each pass. -
- David Green contributed a new --enable-expensive-checks configure - option which enables STL checking, and fixed several bugs exposed by - it. +
- We rewrote the lexer and parser used by TableGen to make them simpler +and cleaner. This gives tblgen support for 'caret diagnostics'. The .ll file +lexer was also rewritten to support caret diagnostics but doesn't use this +support yet. + +
- Dale has been grinding through the GCC testsuite, and marked many +LLVM-incompatible tests as not-to-be-run (for example, if they are grepping +through some GCC dump file that LLVM doesn't produce), he also found and fixed +many LLVM bugs exposed by the testsuite.
- Intel and AMD machines running Red Hat Linux, Fedora Core and FreeBSD (and probably other unix-like systems). -
- PowerPC and X86-based Mac OS X systems, running 10.2 and above in 32-bit and +
- PowerPC and X86-based Mac OS X systems, running 10.3 and above in 32-bit and 64-bit modes. -
- Intel and AMD machines running on Win32 using MinGW libraries (native) +
- Intel and AMD machines running on Win32 using MinGW libraries (native).
- Intel and AMD machines running on Win32 with the Cygwin libraries (limited support is available for native builds with Visual C++).
- Sun UltraSPARC workstations running Solaris 8. @@ -348,13 +397,12 @@ 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 -cee pass is known to be buggy, and may be removed in a - future release. -
- The MSIL backend is experimental. -
- The IA64 code generator is experimental. -
- The Alpha backend is experimental. -
- "-filetype=asm" (the default) is the only supported value for the - -filetype llc option. +
- The -cee pass is known to be buggy and will be removed in + LLVM 2.3. +
- The MSIL, IA64, Alpha, and MIPS backends are experimental. +
- The LLC "-filetype=asm" (the default) is the only supported + value for this option. +
- The llvmc tool is not supported.
-
-
-
-
- PowerPC backend does not correctly -implement ordered FP comparisons.
- The Linux PPC32/ABI support needs testing for the interpreter and static compilation, and lacks support for debug information.
+The llvm-gcc 4.2 Ada compiler works fairly well, however this is not a mature
+technology and problems should be expected.
+
+
+
+
+
+-
+
- The Ada front-end currently only builds on x86-32. This is mainly due +to lack of trampoline support (pointers to nested functions) on other platforms, +however it also fails to build on x86-64 +which does support trampolines. +
- The Ada front-end fails to bootstrap. +Workaround: configure with --disable-bootstrap. +
- The c380004 and c393010 ACATS tests +fail (c380004 also fails with gcc-4.2 mainline). +
- Many gcc specific Ada tests continue to crash the compiler. +
- The -E binder option (exception backtraces) +does not work and will result in programs +crashing if an exception is raised. Workaround: do not use -E. +
- Only discrete types are allowed to start +or finish at a non-byte offset in a record. Workaround: do not pack records +or use representation clauses that result in a field of a non-discrete type +starting or finishing in the middle of a byte. +
- The lli interpreter considers 'main' +as generated by the Ada binder to be invalid. +Workaround: hand edit the file to use pointers for argv and envp rather than +integers. +
- The -fstack-check option is ignored. +
+
+
+
-
+
- The llvm-gcc 4.2 gfortran front-end supports a broad range of Fortran code, but does +not support EQUIVALENCE yet. +
Bugs
-
@@ -515,10 +610,6 @@ llvmdev mailing list if you are interested.
llvm-gcc4 does not currently support Link-Time -Optimization on most platforms "out-of-the-box". Please inquire on the +
llvm-gcc does not currently support Link-Time +Optimization on most platforms "out-of-the-box". Please inquire on the llvmdev mailing list if you are interested.
-
-
"long double" is silently transformed by the front-end into "double". There -is no support for floating point data types of any size other than 32 and 64 -bits.
-
llvm-gcc does not support __builtin_apply yet. See Constructing Calls: Dispatching a call to another function.
@@ -529,7 +620,7 @@ bits.
As in Algol and Pascal, lexical scoping of functions.
Nested functions are supported, but llvm-gcc does not support
- taking the address of a nested function (except on the X86-32 target)
+ taking the address of a nested function (except on X86 targets)
or non-local gotos.
- Function Attributes: @@ -541,11 +632,11 @@ bits. const, constructor, destructor, deprecated, fastcall, format, format_arg, non_null, noinline, - noreturn, pure, regparm + noreturn, nothrow, pure, regparm section, stdcall, unused, used, visibility, warn_unused_result, weak
- Exception handling only works well on the linux/X86-32 target. -In some cases, illegally throwing an exception does not result -in a call to terminate. - - +
- Exception handling only works well on the X86 and PowerPC targets. +It works well for x86-64 darwin but not x86-64 linux.
- Ignored: nothrow, malloc, + Ignored: malloc, no_instrument_function @@ -623,29 +714,8 @@ tested and works for a number of non-trivial programs, including LLVM itself, Qt, Mozilla, etc.
-
-
Last modified: $Date$