Clone this repo:
  1. a36e9d2 Remove space, silence -Wdeprecated-literal-operator. am: 9c7f1121f5 by Vladimir Marko · 8 weeks ago main master
  2. 9c7f112 Remove space, silence -Wdeprecated-literal-operator. by Vladimir Marko · 8 weeks ago
  3. a334f9c Convert run-vixl-tests from to Android.bp am: 4a2aa6b63b by yangbill · 3 months ago
  4. 4a2aa6b Convert run-vixl-tests from to Android.bp by yangbill · 3 months ago
  5. 818ea0b Remove dex2oat-pgo-defaults am: 569113fb31 am: 01a5996828 am: 1e89aaf863 by Dmitrii Ishcheikin · 6 months ago

VIXL: ARMv8 Runtime Code Generation Library


  • Overview
  • Licence
  • Requirements
  • Known limitations
  • Bug reports
  • Usage


VIXL contains three components.

  1. Programmatic assemblers to generate A64, A32 or T32 code at runtime. The assemblers abstract some of the constraints of each ISA; for example, most instructions support any immediate.
  2. Disassemblers that can print any instruction emitted by the assemblers.
  3. A simulator that can simulate any instruction emitted by the A64 assembler. The simulator allows generated code to be run on another architecture without the need for a full ISA model.

The VIXL git repository can be found on GitHub.

Build and Test Status

  • Build Status Simulator
  • Build Status Native
  • Build Status MacOS


This software is covered by the licence described in the LICENCE file.

Contributions, as pull requests or via other means, are accepted under the terms of the same LICENCE.


To build VIXL the following software is required:

  1. Python 2.7
  2. SCons 2.0
  3. GCC 4.8+ or Clang 4.0+

A 64-bit host machine is required, implementing an LP64 data model. VIXL has been tested using GCC on AArch64 Debian, GCC and Clang on amd64 Ubuntu systems.

To run the linter and code formatting stages of the tests, the following software is also required:

  1. Git
  2. Google's
  3. clang-format-4.0
  4. clang-tidy-4.0

Refer to the ‘Usage’ section for details.

Note that in Ubuntu 18.04, clang-tidy-4.0 will only work if the clang-4.0 package is also installed.

Supported Arm Architecture Features

FeatureVIXL CPUFeatures FlagNotes
BTIkBTIPer-page enabling not supported
FP16kFPHalf, kNEONHalf
PAuthkPAuth, kPAuthGenericNot ERETAA, ERETAB

Enable generating code for an architecture feature by combining a flag with the MacroAssembler's defaults. For example, to generate code for SVE, use masm.GetCPUFeatures()->Combine(CPUFeatures::kSVE);.

See the cpu features header file for more information.

Known Limitations

VIXL was developed for JavaScript engines so a number of features from A64 were deemed unnecessary:

  • Limited rounding mode support for floating point.
  • Limited support for synchronisation instructions.
  • Limited support for system instructions.
  • A few miscellaneous integer and floating point instructions are missing.

The VIXL simulator supports only those instructions that the VIXL assembler can generate. The doc directory contains a list of supported A64 instructions.

The VIXL simulator was developed to run on 64-bit amd64 platforms. Whilst it builds and mostly works for 32-bit x86 platforms, there are a number of floating-point operations which do not work correctly, and a number of tests fail as a result.

Debug Builds

Your project's build system must define VIXL_DEBUG (eg. -DVIXL_DEBUG) when using a VIXL library that has been built with debug enabled.

Some classes defined in VIXL header files contain fields that are only present in debug builds, so if VIXL_DEBUG is defined when the library is built, but not defined for the header files included in your project, you will see runtime failures.

Exclusive-Access Instructions

All exclusive-access instructions are supported, but the simulator cannot accurately simulate their behaviour as described in the ARMv8 Architecture Reference Manual.

  • A local monitor is simulated, so simulated exclusive loads and stores execute as expected in a single-threaded environment.
  • The global monitor is simulated by occasionally causing exclusive-access instructions to fail regardless of the local monitor state.
  • Load-acquire, store-release semantics are approximated by issuing a host memory barrier after loads or before stores. The built-in __sync_synchronize() is used for this purpose.

The simulator tries to be strict, and implements the following restrictions that the ARMv8 ARM allows:

  • A pair of load-/store-exclusive instructions will only succeed if they have the same address and access size.
  • Most of the time, cache-maintenance operations or explicit memory accesses will clear the exclusive monitor.
    • To ensure that simulated code does not depend on this behaviour, the exclusive monitor will sometimes be left intact after these instructions.

Instructions affected by these limitations: stxrb, stxrh, stxr, ldxrb, ldxrh, ldxr, stxp, ldxp, stlxrb, stlxrh, stlxr, ldaxrb, ldaxrh, ldaxr, stlxp, ldaxp, stlrb, stlrh, stlr, ldarb, ldarh, ldar, clrex.

Security Considerations

VIXL allows callers to generate any code they want. The generated code is arbitrary, and can therefore call back into any other component in the process. As with any self-modifying code, vulnerabilities in the client or in VIXL itself could lead to arbitrary code generation.

For performance reasons, VIXL‘s Assembler only performs debug-mode checking of instruction operands (such as immediate field encodability). This can minimise code-generation overheads for advanced compilers that already model instructions accurately, and might consider the Assembler’s checks to be redundant. The Assembler should only be used directly where encodability is independently checked, and where fine control over all generated code is required.

The MacroAssembler synthesises multiple-instruction sequences to support some unencodable operand combinations. The MacroAssembler can provide a useful safety check in cases where the Assembler's precision is not required; an unexpected unencodable operand should result in a macro with the correct behaviour, rather than an invalid instruction.

In general, the MacroAssembler handles operands which are likely to vary with user-supplied data, but does not usually handle inputs which are likely to be easily covered by tests. For example, move-immediate arguments are likely to be data-dependent, but register types (e.g. x vs w) are not.

We recommend that all users use the MacroAssembler, using ExactAssemblyScope to invoke the Assembler when specific instruction sequences are required. This approach is recommended even in cases where a compiler can model the instructions precisely, because, subject to the limitations described above, it offers an additional layer of protection against logic bugs in instruction selection.

Bug reports

Bug reports may be made in the Issues section of GitHub, or sent to Please provide any steps required to recreate a bug, along with build environment and host system information.


Running all Tests

The helper script tools/ will build and run every test that is provided with VIXL, in both release and debug mode. It is a useful script for verifying that all of VIXL's dependencies are in place and that VIXL is working as it should.

By default, the tools/ script runs a linter to check that the source code conforms with the code style guide, and to detect several common errors that the compiler may not warn about. This is most useful for VIXL developers. The linter has the following dependencies:

  1. Git must be installed, and the VIXL project must be in a valid Git repository, such as one produced using git clone.
  2., as provided by Google, must be available (and executable) on the PATH.

It is possible to tell tools/ to skip the linter stage by passing --nolint. This removes the dependency on and Git. The --nolint option is implied if the VIXL project is a snapshot (with no .git directory).

Additionally, tools/ tests code formatting using clang-format-4.0, and performs static analysis using clang-tidy-4.0. If you don't have these tools, disable the test using --noclang-format or --noclang-tidy, respectively.

Also note that the tests for the tracing features depend upon external diff and sed tools. If these tools are not available in PATH, these tests will fail.

Getting Started

We have separate guides for introducing VIXL, depending on what architecture you are targeting. A guide for working with AArch32 can be found here, while the AArch64 guide is here. Example source code is provided in the examples directory. You can build examples with either scons aarch32_examples or scons aarch64_examples from the root directory, or use scons --help to get a detailed list of available build targets.