You may be interested in switching to std-simd. GCC 11 includes an experimental version of std::simd as part of libstdc++, which also works with clang. Features present in Vc 1.4 and not present in std-simd will eventually turn into Vc 2.0,which then depends on std-simd.

Vc: portable, zero-overhead C++ types for explicitly data-parallel programming

Recent generations of CPUs, and GPUs in particular, require data-parallel codes for full efficiency. Data parallelism requires that the same sequence of operations is applied to different input data. CPUs and GPUs can thus reduce the necessary hardware for instruction decoding and scheduling in favor of more arithmetic and logic units, which execute the same instructions synchronously. On CPU architectures this is implemented via SIMD registers and instructions. A single SIMD register can store N values and a single SIMD instruction can execute N operations on those values. On GPU architectures N threads run in perfect sync, fed by a single instruction decoder/scheduler. Each thread has local memory and a given index to calculate the offsets in memory for loads and stores.

Current C++ compilers can do automatic transformation of scalar codes to SIMD instructions (auto-vectorization). However, the compiler must reconstruct an intrinsic property of the algorithm that was lost when the developer wrote a purely scalar implementation in C++. Consequently, C++ compilers cannot vectorize any given code to its most efficient data-parallel variant. Especially larger data-parallel loops, spanning over multiple functions or even translation units, will often not be transformed into efficient SIMD code.

The Vc library provides the missing link. Its types enable explicitly stating data-parallel operations on multiple values. The parallelism is therefore added via the type system. Competing approaches state the parallelism via new control structures and consequently new semantics inside the body of these control structures.

Vc is a free software library to ease explicit vectorization of C++ code. It has an intuitive API and provides portability between different compilers and compiler versions as well as portability between different vector instruction sets. Thus an application written with Vc can be compiled for:

AVX and AVX2
SSE2 up to SSE4.2 or SSE4a
Scalar
AVX-512 (Vc 2 development)
NEON (in development)
NVIDIA GPUs / CUDA (research)

After Intel dropped MIC support with ICC 18, Vc 1.4 also removes support for it.

Examples

Usage on Compiler Explorer

Simdize Example
Total momentum and time stepping of std::vector<Particle>
Matrix Example: This uses vertical vectorization which does not scale to different vector sizes. However, the example is instructive to compare it with similar solutions of other languages or libraries.
N-vortex solver showing simdized iteration over many std::vector<float>. Note how important the -march flag is, compared to plain -mavx2 -mfma.

Scalar Product

Let's start from the code for calculating a 3D scalar product using builtin floats:

using Vec3D = std::array<float, 3>;
float scalar_product(Vec3D a, Vec3D b) {
  return a[0] * b[0] + a[1] * b[1] + a[2] * b[2];
}

Using Vc, we can easily vectorize the code using the float_v type:

using Vc::float_v
using Vec3D = std::array<float_v, 3>;
float_v scalar_product(Vec3D a, Vec3D b) {
  return a[0] * b[0] + a[1] * b[1] + a[2] * b[2];
}

The above will scale to 1, 4, 8, 16, etc. scalar products calculated in parallel, depending on the target hardware's capabilities.

For comparison, the same vectorization using Intel SSE intrinsics is more verbose and uses prefix notation (i.e. function calls):

using Vec3D = std::array<__m128, 3>;
__m128 scalar_product(Vec3D a, Vec3D b) {
  return _mm_add_ps(_mm_add_ps(_mm_mul_ps(a[0], b[0]), _mm_mul_ps(a[1], b[1])),
                    _mm_mul_ps(a[2], b[2]));
}

The above will neither scale to AVX, AVX-512, etc. nor is it portable to other SIMD ISAs.

Build Requirements

cmake >= 3.0

C++11 Compiler:

GCC >= 4.8.1
clang >= 3.4
ICC >= 18.0.5
Visual Studio 2015 (64-bit target)

Building and Installing Vc

After cloning, you need to initialize Vc's git submodules:

git submodule update --init

Create a build directory:

$ mkdir build
$ cd build

Call cmake with the relevant options:

$ cmake -DCMAKE_INSTALL_PREFIX=/opt/Vc -DBUILD_TESTING=OFF <srcdir>

Build and install:

$ make -j16
$ make install

Documentation

The documentation is generated via doxygen. You can build the documentation by running doxygen in the doc subdirectory. Alternatively, you can find nightly builds of the documentation at:

Publications

Work on integrating the functionality of Vc in the C++ standard library.

License

Vc is released under the terms of the 3-clause BSD license.

bmanga/Vc