This document contains the release notes for the LLVM compiler
-infrastructure, release 1.5. Here we describe the status of LLVM, including any
-known problems and major improvements from the previous release. The most
-up-to-date version of this document can be found on the LLVM 1.5 web site. If you are
-not reading this on the LLVM web pages, you should probably go there because
-this document may be updated after the release.
+infrastructure, release 2.0. 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.
Note that if you are reading this file from CVS or the main LLVM web page,
+
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 the current or previous releases, see the releases page.
This is the sixth public release of the LLVM Compiler Infrastructure.
+
This is the eleventh public release of the LLVM Compiler Infrastructure.
+Being the first major release since 1.0, this release is different in several
+ways from our previous releases:
-
At this time, LLVM is known to correctly compile a wide range of C and C++
-programs, including the SPEC CPU95 & 2000 suite. It includes bug fixes for
-those problems found since the 1.4 release and a large number of new features
-and enhancements, described below.
+
+
We took this as an opportunity to
+break backwards compatibility with the LLVM 1.x bytecode and .ll file format.
+If you have LLVM 1.9 .ll files that you would like to upgrade to LLVM 2.x, we
+recommend the use of the stand alone llvm-upgrade
+tool (which is included with 2.0). We intend to keep compatibility with .ll
+and .bc formats within the 2.x release series, like we did within the 1.x
+series.
+
There are several significant change to the LLVM IR and internal APIs, such
+ as a major overhaul of the type system, the completely new bitcode file
+ format, etc (described below).
+
We designed the release around a 6 month release cycle instead of the usual
+ 3-month cycle. This gave us extra time to develop and test some of the
+ more invasive features in this release.
+
LLVM 2.0 no longer supports the llvm-gcc3 front-end. Users are required to
+ upgrade to llvm-gcc4. llvm-gcc4 includes many features over
+ llvm-gcc3, is faster, and is much easier to
+ build from source.
+
+
+
Note that while this is a major version bump, this release has been
+ extensively tested on a wide range of software. It is easy to say that this
+ is our best release yet, in terms of both features and correctness. This is
+ the first LLVM release to correctly compile and optimize major software like
+ LLVM itself, Mozilla/Seamonkey, Qt 4.3rc1, kOffice, etc out of the box on
+ linux/x86.
+
-This release includes new native code generators for Alpha, IA-64, and SPARC-V8 (32-bit
-SPARC). These code generators are still beta quality, but are progressing
-rapidly.
-
+
+
Changes to the LLVM IR itself:
+
+
+
+
Integer types are now completely signless. This means that we
+ have types like i8/i16/i32 instead of ubyte/sbyte/short/ushort/int
+ etc. LLVM operations that depend on sign have been split up into
+ separate instructions (PR950). This
+ eliminates cast instructions that just change the sign of the operands (e.g.
+ int -> uint), which reduces the size of the IR and makes optimizers
+ simpler to write.
+
+
Integer types with arbitrary bitwidths (e.g. i13, i36, i42, i1057, etc) are
+ now supported in the LLVM IR and optimizations (PR1043). However, neither llvm-gcc
+ (PR1284) nor the native code generators
+ (PR1270) support non-standard width
+ integers yet.
+
+
'Type planes' have been removed (PR411).
+ It is no longer possible to have two values with the same name in the
+ same symbol table. This simplifies LLVM internals, allowing significant
+ speedups.
+
+
Global variables and functions in .ll files are now prefixed with
+ @ instead of % (PR645).
+
+
The LLVM 1.x "bytecode" format has been replaced with a
+ completely new binary representation, named 'bitcode'. The Bitcode Format brings a
+ number of advantages to the LLVM over the old bytecode format: it is denser
+ (files are smaller), more extensible, requires less memory to read,
+ is easier to keep backwards compatible (so LLVM 2.5 will read 2.0 .bc
+ files), and has many other nice features.
+
+
Load and store instructions now track the alignment of their pointer
+ (PR400). This allows the IR to
+ express loads that are not sufficiently aligned (e.g. due to '#pragma
+ packed') or to capture extra alignment information.
+
+
+
Major new features:
+
+
+
+
A number of ELF features are now supported by LLVM, including 'visibility',
+ extern weak linkage, Thread Local Storage (TLS) with the __thread
+ keyword, and symbol aliases.
+ Among other things, this means that many of the special options needed to
+ configure llvm-gcc on linux are no longer needed, and special hacks to build
+ large C++ libraries like Qt are not needed.
+
+
LLVM now has a new MSIL backend. llc -march=msil will now turn LLVM
+ into MSIL (".net") bytecode. This is still fairly early development
+ with a number of limitations.
+
+
A new llvm-upgrade tool
+ exists to migrates LLVM 1.9 .ll files to LLVM 2.0 syntax.
This release includes a new framework
-for building instruction selectors, which has long been the hardest part of
-building a new LLVM target. This framework handles a lot of the mundane (but
-easy to get wrong) details of writing the instruction selector, such as
-generating efficient code for getelementptr instructions, promoting
-small integer types to larger types (e.g. for RISC targets with one size of
-integer registers), expanding 64-bit integer operations for 32-bit hosts, etc.
-Currently, the X86, PowerPC, Alpha, and IA-64 backends use this framework. The
-SPARC backends will be migrated when time permits.
+
New features include:
+
+
+
Precompiled Headers (PCH) are now supported.
+
+
"#pragma packed" is now supported, as are the various features
+ described above (visibility, extern weak linkage, __thread, aliases,
+ etc).
+
+
Tracking function parameter/result attributes is now possible.
+
+
Many internal enhancements have been added, such as improvements to
+ NON_LVALUE_EXPR, arrays with non-zero base, structs with variable sized
+ fields, VIEW_CONVERT_EXPR, CEIL_DIV_EXPR, nested functions, and many other
+ things. This is primarily to supports non-C GCC front-ends, like Ada.
LLVM 1.5 adds supports for custom and
-target-specific calling conventions. Traditionally, the LLVM code
-generators match the native C calling conventions for a target. This is
-important for compatibility, but is not very flexible. This release allows
-custom calling conventions to be established for functions, and defines three
-target-independent conventions (C call, fast call, and cold call) which may be
-supported by code generators. When possible, the LLVM optimizer promotes C
-functions to use the "fastcc" convention, allowing the use of more efficient
-calling sequences (e.g., parameters are passed in registers in the X86 target).
+
New features include:
-
Targets may now also define target-specific calling conventions, allowing
-LLVM to fully support calling convention altering options (e.g. GCC's
--mregparm flag) and well-defined target conventions (e.g. stdcall and
-fastcall on X86).
+
+
The pass manager has been entirely
+ rewritten, making it significantly smaller, simpler, and more extensible.
+ Support has been added to run FunctionPasses interlaced with
+ CallGraphSCCPasses, we now support loop transformations
+ explicitly with LoopPass, and ModulePasses may now use the
+ result of FunctionPasses.
+
+
LLVM 2.0 includes a new loop rotation pass, which converts "for loops" into
+ "do/while loops", where the condition is at the bottom of the loop.
+
+
The Loop Strength Reduction pass has been improved, and we now support
+ sinking expressions across blocks to reduce register pressure.
+
+
The -scalarrepl pass can now promote unions containing FP values
+ into a register, it can also handle unions of vectors of the same
+ size.
+
+
The [Post]DominatorSet classes have been removed from LLVM and clients
+ switched to use the more-efficient ETForest class instead.
+
+
The ImmediateDominator class has also been removed, and clients have been
+ switched to use DominatorTree instead.
+
+
The predicate simplifier pass has been improved, making it able to do
+ simple value range propagation and eliminate more conditionals. However,
+ note that predsimplify is not enabled by default in llvm-gcc.
The release now includes support for proper tail calls, as
-required to implement languages like Scheme. Tail calls make use of two
-features: custom calling conventions (described above), which allow the code
-generator to emit code for the caller to deallocate its own stack when it
-returns. The second feature is a flag on the call
-instruction, which indicates that the callee does not access the callers
-stack frame (indicating that it is acceptable to deallocate the caller stack
-before invoking the callee). LLVM proper tail calls run on the system stack (as
-do normal calls), supports indirect tail calls, tail calls with arbitrary
-numbers of arguments, tail calls where the callee requires more argument space
-than the caller, etc. The only case not supported are varargs calls, but that
-could be added if desired.
+
+New features include:
-
In order for a front-end to get guaranteed tail call, it must mark functions
-as "fastcc", mark calls with the 'tail' marker, and follow the call with a
-return of the called value (or void). The optimizer and code generator attempt
-to handle more general cases, but the simple case will always work if the code
-generator supports tail calls. Here is a simple example:
+
+
+
LLVM now supports software floating point, which allows LLVM to target
+ chips that don't have hardware FPUs (e.g. ARM thumb mode).
+
+
A new register scavenger has been implemented, which is useful for
+ finding free registers after register allocation. This is useful when
+ rewriting frame references on RISC targets, for example.
+
+
Heuristics have been added to avoid coalescing vregs with very large live
+ ranges to physregs. This was bad because it effectively pinned the physical
+ register for the entire lifetime of the virtual register (PR711).
+
+
Support now exists for very simple (but still very useful)
+ rematerialization the register allocator, enough to move
+ instructions like "load immediate" and constant pool loads.
+
+
Switch statement lowering is significantly better, improving codegen for
+ sparse switches that have dense subregions, and implemented support
+ for the shift/and trick.
-
-fastcc int %bar(int %X, int(double, int)* %FP) { ; fastcc
- %Y = tail call fastcc int %FP(double 0.0, int %X) ; tail, fastcc
- ret int %Y
-}
-
+
LLVM now supports tracking physreg sub-registers and super-registers
+ in the code generator, and includes extensive register
+ allocator changes to track them.
+
+
There is initial support for virtreg sub-registers
+ (PR1350).
+
+
+
+
+Other improvements include:
+
+
+
+
+
Inline assembly support is much more solid that before.
+ The two primary features still missing are support for 80-bit floating point
+ stack registers on X86 (PR879), and
+ support for inline asm in the C backend (PR802).
+
+
DWARF debug information generation has been improved. LLVM now passes
+ most of the GDB testsuite on MacOS and debug info is more dense.
+
+
Codegen support for Zero-cost DWARF exception handling has been added (PR592). It is mostly
+ complete and just in need of continued bug fixes and optimizations at
+ this point. However, support in llvm-g++ is disabled with an
+ #ifdef for the 2.0 release (PR870).
+
+
The code generator now has more accurate and general hooks for
+ describing addressing modes ("isLegalAddressingMode") to
+ optimizations like loop strength reduction and code sinking.
+
+
Progress has been made on a direct Mach-o .o file writer. Many small
+ apps work, but it is still not quite complete.
+
+
+
+
In addition, the LLVM target description format has itself been extended in
+ several ways:
+
+
+
TargetData now supports better target parameterization in
+ the .ll/.bc files, eliminating the 'pointersize/endianness' attributes
+ in the files (PR761).
+
+
TargetData was generalized for finer grained alignment handling,
+ handling of vector alignment, and handling of preferred alignment
+
+
LLVM now supports describing target calling conventions
+ explicitly in .td files, reducing the amount of C++ code that needs
+ to be written for a port.
+
+
-
In LLVM 1.5, the X86 code generator is the only target that has been enhanced
-to support proper tail calls (other targets will be enhanced in future).
-Further, because this support was added very close to the release, it is
-disabled by default. Pass -enable-x86-fastcc to llc to enable it. X86
-support will be enabled by default in the next LLVM release.
The MMX instruction set is now supported through intrinsics.
+
The scheduler was improved to better reduce register pressure on
+ X86 and other targets that are register pressure sensitive.
+
Linux/x86-64 support is much better.
+
PIC support for linux/x86 has been added.
+
The X86 backend now supports the GCC regparm attribute.
+
LLVM now supports inline asm with multiple constraint letters per operand
+ (like "mri") which is common in X86 inline asms.
+
+
+
ARM-specific Code Generator Enhancements:
+
+
+
The ARM code generator is now stable and fully supported.
+
+
There are major new features, including support for ARM
+ v4-v6 chips, vfp support, soft float point support, pre/postinc support,
+ load/store multiple generation, constant pool entry motion (to support
+ large functions), inline asm support, weak linkage support, static
+ ctor/dtor support and many bug fixes.
+
+
Added support for Thumb code generation (llc -march=thumb).
+
+
The ARM backend now supports the ARM AAPCS/EABI ABI and PIC codegen on
+ arm/linux.
+
+
Several bugs were fixed for DWARF debug info generation on arm/linux.
+
+
+
+
PowerPC-specific Code Generator Enhancements:
+
+
+
The PowerPC 64 JIT now supports addressing code loaded above the 2G
+ boundary.
+
+
Improved support for the Linux/ppc ABI and the linux/ppc JIT is fully
+ functional now. llvm-gcc and static compilation are not fully supported
+ yet though.
The -globalopt pass now promotes non-address-taken static globals that are
-only accessed in main to SSA registers.
-
Loops with trip counts based on array pointer comparisons (e.g. "for (i
-= 0; &A[i] != &A[100]; ++i) ...") are optimized better than before,
-which primarily helps iterator-intensive C++ codes.
+
More specific changes include:
-
The code generator now uses information about takes advantage of commutative
-two-address instructions when performing register allocation.
+
+
LLVM no longer relies on static destructors to shut itself down. Instead,
+ it lazily initializes itself and shuts down when llvm_shutdown() is
+ explicitly called.
-
-
-
+
LLVM now has significantly fewer static constructors, reducing startup time.
+
LLVM 2.0 contains a revamp of the type system and several other significant
+internal changes. If you are programming to the C++ API, be aware of the
+following major changes:
Transition code for LLVM 1.0 style varargs was removed from the .ll file
- parser. LLVM 1.0 bytecode files are still supported.
-
+
+
Pass registration is slightly different in LLVM 2.0 (you now need an
+ intptr_t in your constructor), as explained in the Writing an LLVM Pass
+ document.
+
+
ConstantBool, ConstantIntegral and ConstantInt
+ classes have been merged together, we now just have
+ ConstantInt.
+
+
Type::IntTy, Type::UIntTy, Type::SByteTy, ... are
+ replaced by Type::Int8Ty, Type::Int16Ty, etc. LLVM types
+ have always corresponded to fixed size types
+ (e.g. long was always 64-bits), but the type system no longer includes
+ information about the sign of the type. Also, the
+ Type::isPrimitiveType() method now returns false for integers.
+
+
Several classes (CallInst, GetElementPtrInst,
+ ConstantArray, etc), that once took std::vector as
+ arguments now take ranges instead. For example, you can create a
+ GetElementPtrInst with code like:
+
+
+ Value *Ops[] = { Op1, Op2, Op3 };
+ GEP = new GetElementPtrInst(BasePtr, Ops, 3);
+
+
+ This avoids creation of a temporary vector (and a call to malloc/free). If
+ you have an std::vector, use code like this:
+
Instruction::getNext()/getPrev() are now private (along with
+ BasicBlock::getNext, etc), for efficiency reasons (they are now no
+ longer just simple pointers). Please use BasicBlock::iterator, etc
+ instead.
+
Intel and AMD machines running Red Hat Linux and FreeBSD (and probably
- other unix-like systems).
-
Sun UltraSPARC workstations running Solaris 8.
+
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
+ 64-bit modes.
+
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++).
-
PowerPC-based Mac OS X systems, running 10.2 and above.
+
Sun UltraSPARC workstations running Solaris 8.
Alpha-based machines running Debian GNU/Linux.
Itanium-based machines running Linux and HP-UX.
@@ -306,7 +511,7 @@ portability patches and reports of successful builds or error messages.
This section contains all known problems with the LLVM system, listed by
component. As new problems are discovered, they will be added to these
sections. If you run into a problem, please check the LLVM bug database and submit a bug if
+href="http://llvm.org/bugs/">LLVM bug database and submit a bug if
there isn't already one.
@@ -322,169 +527,213 @@ there isn't already one.
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.
+components, please contact us on the LLVMdev list.
-
The following passes are incomplete or buggy, and may be removed in future
- releases: -cee, -branch-combine, -instloops, -paths, -pre
-
The llvm-db tool is in a very early stage of development, but can
- be used to step through programs and inspect the stack.
-
The "iterative scan" register allocator (enabled with
- -regalloc=iterativescan) is not stable.
-
The SparcV8, Alpha, and IA64 code generators are experimental.
+
The -cee pass is known to be buggy, and may be removed in in a
+ future release.
+
C++ EH support is disabled for this release.
+
The MSIL backend is experimental.
+
The IA64 code generator is experimental.
+
The Alpha JIT is experimental.
+
"-filetype=asm" (the default) is the only supported value for the
+ -filetype llc option.
In the JIT, dlsym() on a symbol compiled by the JIT will not
- work.
-
The JIT does not use mutexes to protect its internal data structures. As
- such, execution of a threaded program could cause these data structures to be
- corrupted.
-
On 21164s, some rare FP arithmetic sequences which may trap do not have the
+appropriate nops inserted to ensure restartability.
-
"long double" is 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.
-
-
The following Unix system functionality has not been tested and may not
-work:
-
-
sigsetjmp, siglongjmp - These are not turned into the
- appropriate invoke/unwind instructions. Note that
- setjmp and longjmpare compiled correctly.
-
getcontext, setcontext, makecontext
- - These functions have not been tested.
-
+
+
-
Although many GCC extensions are supported, some are not. In particular,
- the following extensions are known to not be supported:
-
-
The following GCC extensions are partially supported. An ignored
- attribute means that the LLVM compiler ignores the presence of the attribute,
- but the code should still work. An unsupported attribute is one which is
- ignored by the LLVM compiler and will cause a different interpretation of
- the program.
+
-
-
Variable Length:
- Arrays whose length is computed at run time.
- Supported, but allocated stack space is not freed until the function returns (noted above).
C++ programs are likely to fail on IA64, as calls to setjmp are
+made where the argument is not 16-byte aligned, as required on IA64. (Strictly
+speaking this is not a bug in the IA64 back-end; it will also be encountered
+when building C++ programs using the C back-end.)
- Declaring that functions have no side effects or that they can never
- return.
+
The C++ front-end does not use IA64
+ABI compliant layout of v-tables. In particular, it just stores function
+pointers instead of function descriptors in the vtable. This bug prevents
+mixing C++ code compiled with LLVM with C++ objects compiled by other C++
+compilers.
There are a few ABI violations which will lead to problems when mixing LLVM
+output with code built with other compilers, particularly for floating-point
+programs.
- Unsupported:aligned, packed,
- all target specific attributes.
+
-
Other Builtins:
- Other built-in functions.
- We support all builtins which have a C language equivalent (e.g.,
- __builtin_cos), __builtin_alloca,
- __builtin_types_compatible_p, __builtin_choose_expr,
- __builtin_constant_p, and __builtin_expect
- (currently ignored). We also support builtins for ISO C99 floating
- point comparison macros (e.g., __builtin_islessequal),
- __builtin_prefetch, __builtin_popcount[ll],
- __builtin_clz[ll], and __builtin_ctz[ll].
llvm-gcc4 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.
+
+
+
+
+
+ Notes
+
+
+
+
+
+
"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.
+
+
+
llvm-gcc partially supports these GCC extensions:
+
+
Nested Functions: As in Algol and Pascal, lexical scoping of functions.
+ Nested functions are supported, but llvm-gcc does not support non-local
+ gotos or taking the address of a nested function.
+
+
Function Attributes:
+
+ Declaring that functions have no side effects or that they can never
+ return.
+
+ Supported:alias, always_inline, cdecl,
+ constructor, destructor,
+ deprecated, fastcall, format,
+ format_arg, non_null, noreturn, regparm
+ section, stdcall, unused, used,
+ visibility, warn_unused_result, weak
+
+ Ignored:noinline, pure, const, nothrow,
+ malloc, no_instrument_function
+
-
The following extensions are known to be supported:
+
llvm-gcc supports the vast majority of GCC extensions, including:
@@ -534,44 +789,15 @@ lists, please let us know (also including whether or not they work).
-
For this release, the C++ front-end is considered to be fully
+
The C++ front-end is considered to be fully
tested and works for a number of non-trivial programs, including LLVM
-itself.
-
-
-
-
-
Bugs
-
-
-
-
-
The C++ front-end inherits all problems afflicting the C
- front-end.
-
-
IA-64 specific: The C++ front-end does not use IA64 ABI compliant layout of v-tables.
-In particular, it just stores function pointers instead of function
-descriptors in the vtable. This bug prevents mixing C++ code compiled with
-LLVM with C++ objects compiled by other C++ compilers.
-
-
-
-
-
-
-
- Notes
-
-
-
+itself, Qt, Mozilla, etc.
+
llvm-gcc4 only has partial support for C++
+Exception Handling, and it is not enabled by default.
-
The C++ front-end is based on a pre-release of the GCC 3.4 C++ parser. This
-parser is significantly more standards compliant (and picky) than prior GCC
-versions. For more information, see the C++ section of the GCC 3.4 release notes.
The C back-end produces code that violates the ANSI C Type-Based Alias
-Analysis rules. As such, special options may be necessary to compile the code
-(for example, GCC requires the -fno-strict-aliasing option). This
-problem probably cannot be fixed.
On 21164s, some rare FP arithmatic sequences which may trap do not have the appropriate nops inserted to ensure restartability.
-
-
Vararg functions are not supported.
-
-
Due to the vararg problems, C++ exceptions do not work. Small changes are required to the CFE (which break correctness in the exception handler) to compile the exception handling library (and thus the C++ standard library).
C++ programs are likely to fail on IA64, as calls to setjmp are
-made where the argument is not 16-byte aligned, as required on IA64. (Strictly
-speaking this is not a bug in the IA64 back-end; it will also be encountered
-when building C++ programs using the C back-end.)
-
There are a few ABI violations which will lead to problems
-when mixing LLVM output with code built with other compilers,
-particularly for C++ and floating-point programs.
-
-
Vararg functions are not supported.
-
-
-
-
@@ -707,16 +830,16 @@ particularly for C++ and floating-point programs.
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.