1 VIXL: AArch64 Runtime Code Generation Library Version 1.9 2 ========================================================= 3 4 Contents: 5 6 * Overview 7 * Requirements 8 * Known limitations 9 * Usage 10 11 12 Overview 13 ======== 14 15 VIXL contains three components. 16 17 1. A programmatic **assembler** to generate A64 code at runtime. The assembler 18 abstracts some of the constraints of the A64 ISA; for example, most 19 instructions support any immediate. 20 2. A **disassembler** that can print any instruction emitted by the assembler. 21 3. A **simulator** that can simulate any instruction emitted by the assembler. 22 The simulator allows generated code to be run on another architecture 23 without the need for a full ISA model. 24 25 The VIXL git repository can be found [on GitHub][vixl]. 26 27 Changes from previous versions of VIXL can be found in the 28 [Changelog](doc/changelog.md). 29 30 31 Requirements 32 ============ 33 34 To build VIXL the following software is required: 35 36 1. Python 2.7 37 2. SCons 2.0 38 3. GCC 4.8+ or Clang 3.4+ 39 40 A 64-bit host machine is required, implementing an LP64 data model. VIXL has 41 been tested using GCC on AArch64 Debian, GCC and Clang on amd64 Ubuntu 42 systems. 43 44 To run the linter stage of the tests, the following software is also required: 45 46 1. Git 47 2. [Google's `cpplint.py`][cpplint] 48 49 Refer to the 'Usage' section for details. 50 51 52 Known Limitations 53 ================= 54 55 VIXL was developed for JavaScript engines so a number of features from A64 were 56 deemed unnecessary: 57 58 * Limited rounding mode support for floating point. 59 * Limited support for synchronisation instructions. 60 * Limited support for system instructions. 61 * A few miscellaneous integer and floating point instructions are missing. 62 63 The VIXL simulator supports only those instructions that the VIXL assembler can 64 generate. The `doc` directory contains a 65 [list of supported instructions](doc/supported-instructions.md). 66 67 The VIXL simulator was developed to run on 64-bit amd64 platforms. Whilst it 68 builds and mostly works for 32-bit x86 platforms, there are a number of 69 floating-point operations which do not work correctly, and a number of tests 70 fail as a result. 71 72 Debug Builds 73 ------------ 74 75 Your project's build system must define `VIXL_DEBUG` (eg. `-DVIXL_DEBUG`) 76 when using a VIXL library that has been built with debug enabled. 77 78 Some classes defined in VIXL header files contain fields that are only present 79 in debug builds, so if `VIXL_DEBUG` is defined when the library is built, but 80 not defined for the header files included in your project, you will see runtime 81 failures. 82 83 Exclusive-Access Instructions 84 ----------------------------- 85 86 All exclusive-access instructions are supported, but the simulator cannot 87 accurately simulate their behaviour as described in the ARMv8 Architecture 88 Reference Manual. 89 90 * A local monitor is simulated, so simulated exclusive loads and stores execute 91 as expected in a single-threaded environment. 92 * The global monitor is simulated by occasionally causing exclusive-access 93 instructions to fail regardless of the local monitor state. 94 * Load-acquire, store-release semantics are approximated by issuing a host 95 memory barrier after loads or before stores. The built-in 96 `__sync_synchronize()` is used for this purpose. 97 98 The simulator tries to be strict, and implements the following restrictions that 99 the ARMv8 ARM allows: 100 101 * A pair of load-/store-exclusive instructions will only succeed if they have 102 the same address and access size. 103 * Most of the time, cache-maintenance operations or explicit memory accesses 104 will clear the exclusive monitor. 105 * To ensure that simulated code does not depend on this behaviour, the 106 exclusive monitor will sometimes be left intact after these instructions. 107 108 Instructions affected by these limitations: 109 `stxrb`, `stxrh`, `stxr`, `ldxrb`, `ldxrh`, `ldxr`, `stxp`, `ldxp`, `stlxrb`, 110 `stlxrh`, `stlxr`, `ldaxrb`, `ldaxrh`, `ldaxr`, `stlxp`, `ldaxp`, `stlrb`, 111 `stlrh`, `stlr`, `ldarb`, `ldarh`, `ldar`, `clrex`. 112 113 114 Usage 115 ===== 116 117 Running all Tests 118 ----------------- 119 120 The helper script `tools/presubmit.py` will build and run every test that is 121 provided with VIXL, in both release and debug mode. It is a useful script for 122 verifying that all of VIXL's dependencies are in place and that VIXL is working 123 as it should. 124 125 By default, the `tools/presubmit.py` script runs a linter to check that the 126 source code conforms with the code style guide, and to detect several common 127 errors that the compiler may not warn about. This is most useful for VIXL 128 developers. The linter has the following dependencies: 129 130 1. Git must be installed, and the VIXL project must be in a valid Git 131 repository, such as one produced using `git clone`. 132 2. `cpplint.py`, [as provided by Google][cpplint], must be available (and 133 executable) on the `PATH`. 134 135 It is possible to tell `tools/presubmit.py` to skip the linter stage by passing 136 `--nolint`. This removes the dependency on `cpplint.py` and Git. The `--nolint` 137 option is implied if the VIXL project is a snapshot (with no `.git` directory). 138 139 140 Building and Running the Benchmarks 141 ----------------------------------- 142 143 There are three very basic benchmarks provided with VIXL: 144 145 1. bench-dataop, emitting adds 146 2. bench-branch, emitting branches 147 3. bench-branch-link, emitting branch-links 148 149 Build these benchmarks using `scons bench-dataop`, `scons bench-branch` and 150 `scons bench-branch-link`. This will produce binaries called 151 `bench-dataop_sim`, `bench-branch_sim` and `bench-branch-link_sim`. Run these 152 with an iteration count argument, for example `./bench-dataop_sim 10000000`. The 153 benchmarks do not report a result; time them using the UNIX `time` command. 154 155 Build the benchmarks natively for execution on an AArch64 target using `scons 156 <benchmark name> simulator=off`. This will produce binaries called 157 `bench-dataop`, `bench-branch` and `bench-branch-link`. Run and time these in 158 the same way as the simulator versions. 159 160 161 Getting Started 162 --------------- 163 164 A short introduction to using VIXL can be found [here](doc/getting-started.md). 165 Example source code is provided in the [examples](examples) directory. You can 166 build all the examples with `scons examples` from the root directory, or use 167 `scons --help` to get a detailed list of available build targets. 168 169 170 Using VIXL 171 ---------- 172 173 In addition to [getting started](doc/getting-started.md) and the 174 [examples](examples), you can find documentation and guides on various topics 175 that may be helpful [here](doc/topics/index.md). 176 177 178 179 180 181 [cpplint]: http://google-styleguide.googlecode.com/svn/trunk/cpplint/cpplint.py 182 "Google's cpplint.py script." 183 184 [vixl]: https://github.com/armvixl/vixl 185 "The VIXL repository on GitHub." 186