This document contains the release notes for the LLVM Compiler Infrastructure, release 8.0.0. Here we describe the status of LLVM, including major improvements from the previous release, improvements in various subprojects of LLVM, and some of the current users of the code. 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.
As discussed on the mailing list, building LLVM will soon require more recent toolchains as follows:
A new CMake check when configuring LLVM provides a soft-error if your
toolchain will become unsupported soon. You can opt out of the soft-error by
LLVM_TEMPORARILY_ALLOW_OLD_TOOLCHAIN CMake variable to
These are issues that couldn’t be fixed before the release. See the bug reports for the latest status.
- The llvm-cov tool can now export lcov trace files using the -format=lcov option of the export command.
add_llvm_loadable_moduleCMake macro has been removed. The
add_llvm_librarymacro with the
MODULEargument now provides the same functionality. See Writing an LLVM Pass.
- For MinGW, references to data variables that might need to be imported from a dll are accessed via a stub, to allow the linker to convert it to a dllimport if needed.
- Added support for labels as offsets in
- Support for precise identification of X86 instructions with memory operands,
by using debug information. This supports profile-driven cache prefetching.
It is enabled with the
- Support for profile-driven software cache prefetching on X86. This is part of
a larger system, consisting of: an offline cache prefetches recommender,
AutoFDO tooling, and LLVM. In this system, a binary compiled with
-x86-discriminate-memopsis run under the observation of the recommender. The recommender identifies certain memory access instructions by their binary file address, and recommends a prefetch of a specific type (NTA, T0, etc) be performed at a specified fixed offset from such an instruction’s memory operand. Next, this information needs to be converted to the AutoFDO syntax and the resulting profile may be passed back to the compiler with the LLVM flag
-prefetch-hints-file, together with the exact same set of compilation parameters used for the original binary. More information is available in the RFC.
- Windows support for libFuzzer (x86_64).
- Function attribute
speculative_load_hardeninghas been introduced to allow indicating that Speculative Load Hardening must be enabled for the function body.
The ORC (On Request Compilation) JIT APIs have been updated to support concurrent compilation. The existing (non-concurrent) ORC layer classes and related APIs are deprecated, have been renamed with a “Legacy” prefix (e.g. LegacyIRCompileLayer). The deprecated clasess will be removed in LLVM 9.
An example JIT stack using the concurrent ORC APIs, called LLJIT, has been added (see include/llvm/ExecutionEngine/Orc/LLJIT.h). The lli tool has been updated to use LLJIT.
MCJIT and ExecutionEngine continue to be supported, though ORC should be preferred for new projects.
- Support for Speculative Load Hardening has been added.
- Initial support for the Tiny code model, where code and its statically defined symbols must live within 1MB of each other.
- Added support for the
.arch_extensionassembler directive, just like on ARM.
- Added support for Hexagon/HVX V66 ISA.
- Improved support of GlobalISel instruction selection framework.
- Implemented emission of
R_MICROMIPS_JALRrelocations. These relocations provide hints to a linker for optimization of jumps to protected symbols.
- ORC JIT has been supported for MIPS and MIPS64 architectures.
- Assembler now suggests alternative MIPS instruction mnemonics when an invalid one is specified.
- Improved support for MIPS N32 ABI.
- Added new instructions (
- Numerous bug fixes and code cleanups.
- Switched to non-PIC default
- Deprecated Darwin support
- Enabled Out-of-Order scheduling for P9
- Better overload rules for compatible vector type parameter
- Support constraint ‘wi’, modifier ‘x’ and VSX registers in inline asm
- Added new builtins like vector int128
- Provided significant improvements to the automatic vectorizer
- Code-gen improvements (especially for Power9)
- Fixed some long-standing bugs in the back end
- Added experimental prologue/epilogue improvements
- Enabled builtins tests in compiler-rt
floattitfin compiler-rt to support conversion between IBM double-double and unsigned int128
- Disable randomized address space when running the sanitizers on Linux ppc64le
- Completed support in LLD for ELFv2
- Enabled llvm-exegesis latency mode for PPC
- A number of bugs related to C/C++ language vector extension support were
-mzvectoroption now actually enables the
vec_stepintrinsic now works, and the
vec_orcintrinsics now generate correct code.
__float128keyword, which had been accidentally enabled in some earlier releases, is now no longer supported. On SystemZ, the
long doubledata type itself already uses the IEEE 128-bit floating-point format.
- When the compiler inlines
memcmp, the generated code no longer returns
INT_MINas the negative result value under any circumstances.
- Various code-gen improvements, in particular related to improved auto-vectorization, inlining, and instruction scheduling.
- Machine model for AMD bdver2 (Piledriver) CPU was added. It is used to support instruction scheduling and other instruction cost heuristics.
- New AVX512F gather and scatter intrinsics were added that take a <X x i1> mask
instead of a scalar integer. This removes the need for a bitcast in IR. The
new intrinsics are named like the old intrinsics with
llvm.avx512.mask.. The old intrinsics will be removed in a future release.
cascadelakeas a CPU name for -march. This is
skylake-avx512with the addition of the
- ADCX instruction will no longer be emitted. This instruction is rarely better than the legacy ADC instruction and just increased code size.
The WebAssembly target is no longer “experimental”! It’s now built by default, rather than needing to be enabled with LLVM_EXPERIMENTAL_TARGETS_TO_BUILD.
The object file format and core C ABI are now considered stable. That said, the object file format has an ABI versioning capability, and one anticipated use for it will be to add support for returning small structs as multiple return values, once the underlying WebAssembly platform itself supports it. Additionally, multithreading support is not yet included in the stable ABI.
- The Nios2 target was removed from this release.
- Printed source code is now syntax highlighted in the terminal (only for C languages).
- The expression command now supports tab completing expressions.
D is a language with C-like syntax and static typing. It pragmatically combines efficiency, control, and modeling power, with safety and programmer productivity. D supports powerful concepts like Compile-Time Function Execution (CTFE) and Template Meta-Programming, provides an innovative approach to concurrency and offers many classical paradigms.
LDC uses the frontend from the reference compiler combined with LLVM as backend to produce efficient native code. LDC targets x86/x86_64 systems like Linux, OS X, FreeBSD and Windows and also Linux on ARM and PowerPC (32/64 bit). Ports to other architectures like AArch64 and MIPS64 are underway.
Dylan is a multi-paradigm functional and object-oriented programming language. It is dynamic while providing a programming model designed to support efficient machine code generation, including fine-grained control over dynamic and static behavior. Dylan also features a powerful macro facility for expressive metaprogramming.
The Open Dylan compiler can use LLVM as one of its code-generating back-ends, including full support for debug info generation. (Open Dylan generates LLVM bitcode directly using a native Dylan IR and bitcode library.) Development of a Dylan debugger and interactive REPL making use of the LLDB libraries is in progress.
Zig is a system programming language intended to be an alternative to C. It provides high level features such as generics, compile time function execution, and partial evaluation, while exposing low level LLVM IR features such as aliases and intrinsics. Zig uses Clang to provide automatic import of .h symbols, including inline functions and simple macros. Zig uses LLD combined with lazily building compiler-rt to provide out-of-the-box cross-compiling for all supported targets.