The SIMDe header-only library provides fast, portable implementations of SIMD intrinsics on hardware which doesn't natively support them, such as calling SSE functions on ARM. There is no performance penalty if the hardware supports the native implementation (e.g., SSE/AVX runs at full speed on x86, NEON on ARM, etc.).
This makes porting code to other architectures much easier in a few key ways:
First, instead of forcing you to rewrite everything for each architecture, SIMDe lets you get a port up and running almost effortlessly. You can then start working on switching the most performance-critical sections to native intrinsics, improving performance gradually. SIMDe lets (for example) SSE/AVX and NEON code exist side-by-side, in the same implementation.
Second, SIMDe makes it easier to write code targeting ISA extensions you don't have convenient access to. You can run NEON code on your x86 machine without an emulator. Obviously you'll eventually want to test on the actual hardware you're targeting, but for most development, SIMDe can provide a much easier path.
SIMDe takes a very different approach from most other SIMD abstraction layers in that it aims to expose the entire functionality of the underlying instruction set. Instead of limiting functionality to the lowest common denominator, SIMDe tries to minimize the amount of effort required to port while still allowing you the space to optimize as needed.
The current focus is on writing complete portable implementations, though a large number of functions already have accelerated implementations using one (or more) of the following:
- SIMD intrinsics from other ISA extensions (e.g., using NEON to implement SSE).
- Compiler-specific vector extensions and built-ins such as
__builtin_shufflevectorand__builtin_convertvector - Compiler auto-vectorization hints, using:
You can try SIMDe online using Compiler Explorer and an amalgamated SIMDe header.
If you have any questions, please feel free to use the issue tracker or the mailing list.
There are currently complete implementations of the following instruction set extensions:
- x86 / x86_64
- WebAssembly
As well as partial support for many others, including NEON and SVE in addition to several AVX-512 extensions. See the instruction-set-support label in the issue tracker for details on progress. If you'd like to be notified when an instruction set is available you may subscribe to the relevant issue.
If you have a project you're interested in using with SIMDe but we don't yet support all the functions you need, please file an issue with a list of what's missing so we know what to prioritize.
The default branch is protected so commits never reach it unless they have passed extensive CI checks. Status badges don't really make sense since they will always be green, but here are the links:
- GitHub Actions
- Cirrus CI
- Semaphore CI
- Circle CI
- AppVeyor
- Azure Pipelines
- Drone CI
- Travis CI
- Packit CI
If you're adding a new build I suggest Cirrus CI, which is where we currently have the most room given the number of builds currently on the platform and the quotas for free/open-source usage. Alternately, feel free to set up another provider (such as Codefresh, Shippable, Bitrise, Werkaer, etc.).
Notice: we plan on changing the name of the default branch from "master" to something else soon; we are just trying to wait to see what name git settles on so we can be consistent.
First off, if you're reading this: thank you! Even considering contributing to SIMDe is very much appreciated!
SIMDe is a fairly large undertaking; there are a lot of functions to get through and a lot of opportunities for optimization on different platforms, so we're very happy for any help you can provide.
Programmers of all skill levels are welcome, there are lots of tasks which are pretty straightforward and don't require any special expertise.
If you're not sure how you'd like to contribute, please consider taking a look at the issue tracker. There is a good first issue tag if you want to ease into a your first contributions, but if you're interested in something else please get in touch via the issue tracker; we're happy to help you get a handle on whatever you are interested in.
If you're interested in implementing currently unimplemented functions, there is a guide explaining how to add new functions and how to quickly and easily get a test case in place. It's a bit rough right now, but if anything is unclear please feel free to use the issue tracker to ask about anything you're not clear on.
First, it is important to note that you do not need two separate
versions (one using SIMDe, the other native). If the native functions
are available SIMDe will use them, and compilers easily optimize away
any overhead from SIMDe; all they have to do is some basic inlining.
-O2 should be enough, but we strongly recommend -O3 (or whatever
flag instructs your compiler to aggressizely optimize) since many of
the portable fallbacks are substantially faster with aggressive
auto-vectorization that isn't enabled at lower optimization levels.
Each instruction set has a separate file; x86/mmx.h for MMX,
x86/sse.h for SSE, x86/sse2.h for SSE2, and so on. Just include
the header for whichever instruction set(s) you want instead of the
native version (if you include the native version after SIMDe it will
result in compile-time errors if native aliases are enabled). SIMDe
will provide the fastest implementation it can given which extensions
you've enabled in your compiler (i.e., if you want to use NEON to
implement SSE, you may need to pass something like -mfpu=neon
or -march=armv8-a+simd. See
GCC ARM-Options
for more information).
If you define SIMDE_ENABLE_NATIVE_ALIASES before including SIMDe
you can use the same names as the native functions. Unfortunately,
this is somewhat error-prone due to portability issues in the APIs, so
it's recommended to only do this for testing. When
SIMDE_ENABLE_NATIVE_ALIASES is undefined only the versions prefixed
with simde_ will be available; for example, the MMX _mm_add_pi8
intrinsic becomes simde_mm_add_pi8, and __m64 becomes simde__m64.
Since SIMDe is meant to be portable, many functions which assume types
are of a specific size have been altered to use fixed-width types
instead. For example, Intel's APIs use char for signed 8-bit
integers, but char on ARM is generally unsigned. SIMDe uses int8_t
to make the API portable, but that means your code may require some
minor changes (such as using int8_t instead of char) to work on
other platforms.
That said, the changes are usually quite minor. It's often enough to just use search and replace, manual changes are required pretty infrequently.
SIMDe makes extensive use of annotations to help the compiler vectorize code. By far the best annotations use the SIMD support built in to OpenMP 4, so if your compiler supports these annotations we strongly recommend you enable them.
If you are already using OpenMP, SIMDe will automatically detect it
using the _OPENMP macro and no further action is required.
Some compilers allow you to enable OpenMP SIMD without enabling the
full OpenMP. In such cases there is no runtime dependency on OpenMP
and no runtime overhead; SIMDe will just be faster. Unfortunately,
SIMDe has no way to detect such situations (the _OPENMP macro is not
defined), so after enabling it in your compiler you'll need to define
SIMDE_ENABLE_OPENMP (e.g., by passing -DSIMDE_ENABLE_OPENMP) to get
SIMDe to output the relevant pragmas.
Enabling OpenMP SIMD support varies by compiler:
- GCC 4.9+ and clang 6+ support a
-fopenmp-simdcommand line flag. - ICC supports a
-qopenmp-simdcommand line flag. - MCST's LCC enables OpenMP SIMD by default, so no flags are needed
(technically you don't even need to pass
-DSIMDE_ENABLE_OPENMP).
We are not currently aware of any other compilers which allow you to enable OpenMP SIMD support without enabling full OpenMP (if you are please file an issue to let us know). You should determine whether you wish to enable full OpenMP support on a case-by-case basis, but it is likely that the overhead of linking to (but not using) the OpenMP runtime library will be dwarfed by the performance improvements from using the OpenMP SIMD annotations in SIMDe.
If you choose not to use OpenMP SIMD, SIMDe also supports using Cilk Plus, GCC loop-specific pragmas, or clang pragma loop hint directives, though these are not nearly as effective as OpenMP SIMD and depending on them will likely result in less efficient code. All of these are detected automatically by SIMDe, so if they are enabled in your compiler nothing more is required.
If for some reason you do not wish to enable OpenMP 4 SIMD support even
though SIMDe detects it, you should define SIMDE_DISABLE_OPENMP prior
to including SIMDe.
SIMDe does depend on some C99 features, though the subset supported by MSVC also works. While we do our best to make sure we provide optimized implementations where they are supported, SIMDe does contain portable fallbacks which are designed to work on any C99 compiler.
Every commit is tested in CI on multiple compilers, platforms, and configurations, and our test coverage is extremely extensive. Currently tested compilers include:
- GCC versions back to 4.8
- Clang versions back to 3.8
- Microsoft Visual Studio back to 12 (2013)
- IBM XL C/C++
- Intel C/C++ Compiler (ICC)
I'm generally willing to accept patches to add support for other compilers, as long as they're not too disruptive, especially if we can get CI support going. If using one of our existing CI providers isn't an option then other CI platforms can be added.
The following architectures are tested in CI for every commit:
- x86_64
- x86
- AArch64
- ARMv8
- ARMv7
- PPC64
- MIPS Loongson
We would love to add more, so patches are extremely welcome!
- The "builtins" module in
portable-snippets
does much the same thing, but for compiler-specific intrinsics
(think
__builtin_clzand_BitScanForward), not SIMD intrinsics. - Intel offers an emulator, the Intel® Software Development Emulator which can be used to develop software which uses Intel intrinsics without having to own hardware which supports them, though it doesn't help for deployment.
- Iris is the only other project I'm aware of which is attempting to create portable implementations like SIMDe. SIMDe is much further along on the Intel side, but Iris looks to be in better shape on ARM. C++-only, Apache 2.0 license. AFAICT there are no accelerated fallbacks, nor is there a good way to add them since it relies extensively on templates.
- There are a few projects trying to implement one set with another:
- ARM_NEON_2_x86_SSE — implementing NEON using SSE. Quite extensive, Apache 2.0 license.
- sse2neon — implementing SSE using NEON. This code has already been merged into SIMDe.
- veclib — implementing SSE2 using AltiVec/VMX, using a non-free IBM library called powerveclib
- SSE-to-NEON — implementing SSE with NEON. Non-free, C++.
- arm-neon-tests contains tests to verify NEON implementations.
If you know of any other related projects, please let us know!
Sometime features can't be emulated. If SIMDe is operating in native mode the functions will work as expected, but if there is no native support some caveats apply:
- Many functions require <math.h> and/or <fenv.h>. SIMDe will still work without those headers, but the results of those functions are undefined.
- x86 / x86_64
- SSE
SIMDE_MM_SET_ROUNDING_MODE()will usefesetround(), altering the global rounding mode.simde_mm_getcsrandsimde_mm_setcsronly implement bits 13 and 14 (rounding mode).
- AVX
simde_mm256_test*do not set the CF/ZF registers as there is no portable way to implement that functionality.simde_mm256_zeroallandsimde_mm256_zeroupperare not implemented as there is no portable way to implement that functionality.
- SSE
Additionally, there are some known limitations which apply when using
native aliases (SIMDE_ENABLE_NATIVE_ALIASES):
- On Windows x86 (but not x86_64), some MMX functions and SSE/SSE2 functions which use MMX types (__m64) other than for pointers may return incorrect results.
Also, as mentioned earlier, while some APIs make assumptions about
basic types (e.g., int is 32 bits), SIMDe does not, so many types
have been altered to use portable fixed-width versions such as
int32_t.
If you find any other differences, please file an issue so we can either fix it or add it to the list above.
SIMDe uses resources provided for free by a number of organizations. While this shouldn't be taken to imply endorsement of SIMDe, we're tremendously grateful for their support:
- IntegriCloud — provides access to a very fast POWER9 server for developing AltiVec/VMX support.
- GCC Compile Farm — provides access to a wide range of machines with different architectures for developing support for various ISA extensions.
- CodeCov.io — provides code coverage analysis for our test cases.
- Google — financing Summer of Code, substantial amounts of code (Sean Maher's contributions), and an Open Source Peer Bonus.
Without such organizations donating resources, SIMDe wouldn't be nearly as useful or usable as it is today.
We would also like to thank anyone who has helped develop the myriad of software on which SIMDe relies, including compilers and analysis tools.
Finally, a special thank you to anyone who has contributed to SIMDe, filed bugs, provided suggestions, or helped with SIMDe development in any way.
SIMDe is distributed under an MIT-style license; see COPYING for details.
Thanks goes to these wonderful people (emoji key):
This project follows the all-contributors specification. Contributions of any kind are welcome!