About vectorize

vectorize is a library for generating reasonably good vectorized code without manually writing intrinsics, handling loop boundaries, etc. It cannot handle all classes of algorithms, only those that can be represented by a for loop with no dependencies on any values other than that at the current index, like taking the square root of every value in an array. In the future it will also represent reductions, like adding up every value in an array or finding the maximum.

It is not an attempt to compete with hand-rolled functions. It is meant to enable some of the "low-hanging fruit" to be vectorized very easily. Often the choice is to quickly write a plain loop in C, or to spend some time writing intrinsics to vectorize it. vectorize presents a third option: write a loop that is as simple to write and read as the C version, but whose performance is much closer to the hand-rolled version.

Examples

apply(n, inputs, outputs, sqrt(_x));
apply(n, inputs, outputs, max(_x, 1.0f) - 1.0f);

Notice how the code is about as easy to read as the C equivalent:

for (unsigned i=0; i<n; ++i)
  outputs[i] = sqrt(inputs[i]);
for (unsigned i=0; i<n; ++i)
  outputs[i] = max(inputs[i], 1.0f) - 1.0f;

Short introduction

The core of vectorize is the apply function. It is similar to the STL algorithm transform. Its function signature is:

template <class F>
void apply(unsigned n, const float* src, float* target, F f);

The trick is that f is an expression whose type encodes the computation. I'll call it the "computation kernel." For example, sqrt(_x) doesn't compute anything, it just generates a type that knows how to compute the square root of vectors and scalars. The apply function knows how to process vectors of data using this computation kernel, and how to handle the remainder as scalars if the size isn't evenly divisible by the vector width.

License

Copyright (c) 2012, Aaron Wishnick All rights reserved.

Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met:

• Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer.
• Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution.
• Neither the name of the nor the names of its contributors may be used to endorse or promote products derived from this software without specific prior written permission.

THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.