This document contains the release notes for the LLVM compiler infrastructure, release 1.3. Here we describe the status of LLVM, including any known problems and bug fixes from the previous release. The most up-to-date version of this document can be found on the LLVM 1.3 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.

For more information about LLVM, including information about potentially more current releases, please check out the main 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 CVS, 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 fourth public release of the LLVM compiler infrastructure. This release primarily improves the performance of the code produced by all aspects of the LLVM compiler, adds many new features, fixes a few bugs, speeds up the compiler, and introduces a new (experimental) PowerPC code generator.

At this time, LLVM is known to correctly compile and run all C & C++ SPEC CPU95 & 2000 benchmarks, the Olden benchmarks, and the Ptrdist benchmarks, and many other programs. LLVM now also works with a broad variety of C++ programs.

1. The LLVM select instruction is now fully implemented and supported by all transformations, native code generators, and the interpreter.
2. Bugpoint can now narrow down code-generation bugs to a loop nest, where before it could only narrow them down to a function being miscompiled.
3. Bugpoint can now debug arbitrary modes of llc and lli, by passing them command line flags (e.g. -regalloc=linearscan).
4. The Control Flow Graph in the native code generators is no longer constrained to be the same as the CFG for the LLVM input code.
5. The LLVM induction variable analysis routines have been rewritten.
6. LLVM now has new loop unrolling and loop unswitching passes.
7. The induction variable substitution pass performs linear function test replacement and exit value replacement optimizations.
8. LLVM now has first-class support for Accurate Garbage Collection, enabling the use of aggressive copying and generational collectors.
9. LLVM now includes a simple implementation of Andersen's interprocedural alias analysis algorithm. NOTE: anders-aa is unfortunately completely broken in LLVM 1.3. See the link for a trivial patch to fix it.
10. Bugpoint can extract individual basic blocks to track down reduce miscompilation testcases.
11. LLVM and the C front-end now work under Win32 using the Cygwin runtime libraries. This includes the JIT compiler.
12. The LLVM code generator is now being documented.
13. LLVM includes a new tool, llvm-bcanalyzer, This tool can compute various statistics and dump information about LLVM bytecode encoding.
14. The LLVM bytecode file format is now documented.
15. LLVM now provides an llvm.isunordered intrinsic for efficient implementation of unordered floating point comparisons.
16. The llvmgcc front-end now supports the GCC builtins for ISO C99 floating point comparison macros (e.g., __builtin_islessequal).
17. We now generate HTML documentation and man pages for the tools from a single source (perl-style POD files).
18. The LLVM code generator can now dynamically load targets from shared objects.
19. LLVM now includes a "skeleton" target, which makes it easier to get started porting LLVM to new architectures.
20. The linear scan register allocator is now enabled by default in the target-independent code generator.
21. LLVM now includes a dead store elimination pass.
22. Bugpoint can now debug miscompilations that lead to the program going into an infinite loop.
23. LLVM now provides interfaces to support ML-style pattern matching on the LLVM IR.
24. LLVM now includes a context-sensitive mod/ref analysis for global variables, which is now enabled by default in gccld.
25. LLVM can now autogenerate assembly printers for code generators from the tablegen description of the target (before they were hand coded).
26. All LLVM tools will now respond to the --version option which will tell you the version of LLVM on which the tool is based.
27. An experimental PowerPC backend has been added, capable of compiling several SPEC benchmarks.

1. LLVM cannot handle structures with more than 256 elements
2. [bugpoint] External functions used in non-instruction entities, such as global constant initializer
3. Stacker does not handle targets with 64-bit pointers.
4. Bugpoint doesn't support uses of external fns by immediate constant exprs
5. Can't add function passes that depend on immutable passes to the FunctionPassManager.
6. Archive file reader doesn't understand abbreviated names in headers

1. LLVM tools will happily spew bytecode onto your terminal
2. [llvmgcc] type names are not emitted for structure typedefs
3. All documentation is now conformant to the HTML 4.01 (Strict) level.
4. The spurious "WARNING: Found global types that are not compatible" warning produced when linking C++ programs has been fixed.
5. lli Doesn't Handle Exceptions From Bytecode Reader
6. Global Vars Have (Somewhat) Limited Type Range
7. operator<< on a Value* now prints the address of the object instead of its contents.
8. Bytecode Enhancements Needed
9. [loopsimplify] Loop simplify is really slow on 252.eon
10. [code-cleanup] SymbolTable class cleanup, Type should not derive from Value, eliminate ConstantPointerRef class.
11. The memory footprint of the LLVM IR has been reduced substantially.
12. The LLVM linker and many core classes have been sped up substantially.

1. Minor configure bugs with -disable/enable-povray and -disable-spec
2. shell scripts output by gccld don't work if you change PATH
3. [llvmgcc] llvmgcc does not compile with gcc 3.4
4. [llvmgcc] obstack.h relies on obsolete casts-as-lvalues GCC extension

1. Fixed: [vmcore] Code quality problem due to long operand of getelementptr
2. The X86 backend now generates substantially better code for 64-bit integer and floating point operations.
3. The -inline pass no longer inlines mutually recursive functions until it hits the inlining threshold.
4. The -inline pass no longer misses obvious inlining opportunities just because the callee eventually calls into an external function.
5. The -simplifycfg pass can now "if convert" simple statements into the select instruction.
6. The -loopsimplify pass can now break natural loops with multiple backedges into multiple nested loops. This enables a variety of subsequent optimizations.
7. The -adce pass can now eliminate calls to functions that do not not write to memory.
8. The link-time optimizer now runs the -prune-eh pass (to remove unused exception handlers).
9. The link-time optimizer now runs dead store elimination and uses a simple interprocedural alias analysis.
10. The -simplifycfg pass can now eliminate simple correlated branches (such as "if (A < B && A < B)", and can turn short-circuiting operators into the strict versions when useful (such as "if (A < B || A > C)" into "if (A < B | A > C)"
11. LLVM now has infrastructure for (simple and sparse conditional) constant propagation of function calls. It currently supports a few math library functions like sqrt/sin/cos/etc.
12. The C backend now emits syntactic loops in the code to help C compilers whose optimizers do not recognize loops formed from gotos (like GCC).
13. The SparcV9 backend no longers spills the null constant to the constant pool.

Bugs fixed in the LLVM Core:

1. [loopsimplify] Loop simplify incorrectly updates dominator information
2. [tailduplicate] DemoteRegToStack breaks SSA form
3. [X86] JIT miscompiles unsigned short to floating point cast
4. [vmcore] Linker causes erroneous asssertion
5. [adce] Crash handling unreachable code that unwinds
6. [sparcv9] LLC can't emit 2 functions of the same name, both having constant pools
7. [livevar] Live variables missed physical register use of aliased definition
8. [X86] stackifier crash on floating point setcc X, X

Bugs in the C/C++ front-end:

1. [llvmgcc] Variable length array indexing miscompiled
2. [llvmgcc] Crash on use of undeclared enum type
3. [llvmgcc] Errors handling function prototypes that take opaque structs by-value
4. [llvmgcc] Crash compiling variable length array of structures
5. [llvmgcc] miscompilation of staticly initialized unsigned bitfields
6. [llvm-gcc] Crash casting function to void

LLVM is known to work in the following platforms:

• 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 on Win32 with the Cygwin libraries.
• PowerPC-based Mac OS X boxes, running 10.2 and above. Note that no JIT support is available yet, and LLC support is beta. The C backend can be used to produce stable code for this platform.

The core LLVM infrastructure uses GNU autoconf to adapt itself to the machine and operating system on which it is built. However, minor porting may be required to get LLVM to work on new platforms. We welcome your portability patches and reports of successful builds or error messages.

Note that the LLVM build system does not currently support directories with spaces on them when running on Win32/cygwin. We strongly recommend running LLVM and the C frontend out of a top-level directory without spaces (e.g., /cygdrive/c/llvm). Also, make sure to install all of the cygwin packages. By default, many important tools are not installed that are needed by the LLVM build process or test suite (e.g., /bin/time). Finally, please make sure that there are no directories with spaces in them in your PATH environment variable.

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 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 PowerPC backend is incomplete and is known to miscompile several SPEC benchmarks. The file llvm/lib/Target/PowerPC/README.txt has details.
• The following passes are incomplete or buggy: -pgmdep, -memdep, -ipmodref, -cee
• The -pre pass is incomplete (there are cases it doesn't handle that it should) and not thoroughly tested.
• The llvm-ar tool is incomplete and probably buggy.
• The llvm-db tool is in a very early stage of development.

• 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.
• Linking in static archive files (.a files) is slow (there is no symbol table in the archive).
• The gccld program does not link objects/archives in the order specified on the command line.
• The lower-invoke pass does not mark values live across a setjmp as volatile. This missing feature only affects targets whose setjmp/longjmp libraries do not save and restore the entire register file.

• [anders-aa] Andersen's AA is completely broken in LLVM 1.3
• [licm] LICM invalidates alias analysis info and uses broken information (optimizer crash)

• C99 Variable sized arrays do not release stack memory when they go out of scope. Thus, the following program may run out of stack space:

      for (i = 0; i != 1000000; ++i) {
        int X[n];
        foo(X);
      }
  

• Initialization of global union variables can only be done with the largest union member.
• [llvm-gcc] Error when an implicitly external function is re-declared as static

• [llvmg++] not enough templates are instantiated
• [llvmg++] Extern const globals cannot be marked 'constant' if they have nontrivial ctors or dtors
• [llvmgcc] Crash compiling unnamed bitfield which does not increase struct size

• Inline assembly is not yet supported.
• "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:
  1. sigsetjmp, siglongjmp - These are not turned into the appropriate invoke/unwind instructions. Note that setjmp and longjmp are compiled correctly.
  2. 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:
  1. Local Labels: Labels local to a block.
  2. Nested Functions: As in Algol and Pascal, lexical scoping of functions.
  3. Constructing Calls: Dispatching a call to another function.
  4. Extended Asm: Assembler instructions with C expressions as operands.
  5. Constraints: Constraints for asm operands.
  6. Asm Labels: Specifying the assembler name to use for a C symbol.
  7. Explicit Reg Vars: Defining variables residing in specified registers.
  8. Vector Extensions: Using vector instructions through built-in functions.
  9. Target Builtins: Built-in functions specific to particular targets.
  10. Thread-Local: Per-thread variables.
  11. Pragmas: Pragmas accepted by GCC.

  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.

  1. Variable Length: Arrays whose length is computed at run time.
     Supported, but allocated stack space is not freed until the function returns (noted above).
  2. Function Attributes: Declaring that functions have no side effects or that they can never return.
     Supported: format, format_arg, non_null, constructor, destructor, unused, deprecated, warn_unused_result, weak
     Ignored: noreturn, noinline, always_inline, pure, const, nothrow, malloc, no_instrument_function, cdecl
     Unsupported: used, section, alias, visibility, regparm, stdcall, fastcall, all other target specific attributes
  3. Variable Attributes: Specifying attributes of variables.
     Supported: cleanup, common, nocommon, deprecated, transparent_union, unused, weak
     Unsupported: aligned, mode, packed, section, shared, tls_model, vector_size, dllimport, dllexport, all target specific attributes.
  4. Type Attributes: Specifying attributes of types.
     Supported: transparent_union, unused, deprecated, may_alias
     Unsupported: aligned, packed, all target specific attributes.
  5. 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).

  The following extensions are known to be supported:

  1. Labels as Values: Getting pointers to labels and computed gotos.
  2. Statement Exprs: Putting statements and declarations inside expressions.
  3. Typeof: typeof: referring to the type of an expression.
  4. Lvalues: Using ?:, "," and casts in lvalues.
  5. Conditionals: Omitting the middle operand of a ?: expression.
  6. Long Long: Double-word integers.
  7. Complex: Data types for complex numbers.
  8. Hex Floats:Hexadecimal floating-point constants.
  9. Zero Length: Zero-length arrays.
  10. Empty Structures: Structures with no members.
  11. Variadic Macros: Macros with a variable number of arguments.
  12. Escaped Newlines: Slightly looser rules for escaped newlines.
  13. Subscripting: Any array can be subscripted, even if not an lvalue.
  14. Pointer Arith: Arithmetic on void-pointers and function pointers.
  15. Initializers: Non-constant initializers.
  16. Compound Literals: Compound literals give structures, unions, or arrays as values.
  17. Designated Inits: Labeling elements of initializers.
  18. Cast to Union: Casting to union type from any member of the union.
  19. Case Ranges: `case 1 ... 9' and such.
  20. Mixed Declarations: Mixing declarations and code.
  21. Function Prototypes: Prototype declarations and old-style definitions.
  22. C++ Comments: C++ comments are recognized.
  23. Dollar Signs: Dollar sign is allowed in identifiers.
  24. Character Escapes: \e stands for the character <ESC>.
  25. Alignment: Inquiring about the alignment of a type or variable.
  26. Inline: Defining inline functions (as fast as macros).
  27. Alternate Keywords:__const__, __asm__, etc., for header files.
  28. Incomplete Enums: enum foo;, with details to follow.
  29. Function Names: Printable strings which are the name of the current function.
  30. Return Address: Getting the return or frame address of a function.
  31. Unnamed Fields: Unnamed struct/union fields within structs/unions.
  32. Attribute Syntax: Formal syntax for attributes.

If you run into GCC extensions which have not been included in any of these lists, please let us know (also including whether or not they work).

For this release, the C++ front-end is considered to be fully functional but has not been tested as thoroughly as the C front-end. It has been tested and works for a number of non-trivial programs, but there may be lurking bugs. Please report any bugs or problems.

• 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.

• [llvmg++] Tons of warnings are spewed when linking to libstdc++

• 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.
• Destructors for local objects are not always run when a longjmp is performed. In particular, destructors for objects in the longjmping function and in the setjmp receiver function may not be run. Objects in intervening stack frames will be destroyed, however (which is better than most compilers).
• The LLVM C++ front-end follows the Itanium C++ ABI. This document, which is not Itanium specific, specifies a standard for name mangling, class layout, v-table layout, RTTI formats, and other C++ representation issues. Because we use this API, code generated by the LLVM compilers should be binary compatible with machine code generated by other Itanium ABI C++ compilers (such as G++, the Intel and HP compilers, etc). However, the exception handling mechanism used by LLVM is very different from the model used in the Itanium ABI, so exceptions will not interact correctly.

• [X86] llc output for functions w/certain names tickles GNU 'as' bugs

• [sparcv9] SparcV9 backend miscompiles several programs in the LLVM test suite
• [sparcv9] regalloc assertion failure with certain indirect calls

• 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.
• Initializers for global variables cannot include special floating point numbers like Not-A-Number or Infinity.
• Zero arg vararg functions are not supported. This should not affect LLVM produced by the C or C++ frontends.

A wide variety of additional information is available on the LLVM web page, including mailing lists and publications describing algorithms and components implemented in LLVM. The web page also contains versions of the API documentation which is up-to-date with the CVS 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.