Writing vectorized code¶
Assume that we have a simple function that computes the mean of two vectors, something like:
#include <cstddef>
#include <vector>
void mean(const std:vector<double>& a, const std::vector<double>& b, std::vector<double>& res)
{
std::size_t size = res.size();
for(std::size_t i = 0; i < size; ++i)
{
res[i] = (a[i] + b[i]) / 2;
}
}
How can we used xsimd to take advantage of vectorization ?
Explicit use of an instruction set¶
xsimd provides the template class batch<T, N>
where N
is the number of scalar values of type T
involved in SIMD
instructions. If you know which intruction set is available on your machine, you can directly use the corresponding specialization
of batch
. For instance, assuming the AVX instruction set is available, the previous code can be vectorized the following way:
#include <cstddef>
#include <vector>
#include "xsimd/xsimd.hpp"
void mean(const std::vector<double>& a, const std::vector<double>& b, std::vector<double>& res)
{
using b_type = xsimd::batch<double, 4>;
std::size_t inc = b_type::size;
std::size_t size = res.size();
// size for which the vectorization is possible
std::size_t vec_size = size - size % inc;
for(std::size_t i = 0; i < vec_size; i +=inc)
{
b_type avec(&a[i]);
b_type bvec(&b[i]);
b_type rvec = (avec + bvec) / 2;
rvec.store_unaligned(&res[i]);
}
// Remaining part that cannot be vectorize
for(std::size_t i = vec_size; i < size; ++i)
{
res[i] = (a[i] + b[i]) / 2;
}
}
However, if you want to write code that is portable, you cannot rely on the use of batch<double, 4>
.
Indeed this won’t compile on a CPU where only SSE2 instruction set is available for instance. To solve this, xsimd provides an
auto-detection mechanism so you can use the most performant SIMD instruction set available on your hardware.
Auto detecting the instruction set¶
Using the auto detection mechanism does not require a lot of change:
#include <cstddef>
#include <vector>
#include "xsimd/xsimd.hpp"
void mean(const std::vector<double>& a, const std::vector<double>& b, std::vector<double>& res)
{
using b_type = xsimd::simd_type<double>;
std::size_t inc = b_type::size;
std::size_t size = res.size();
// size for which the vectorization is possible
std::size_t vec_size = size - size % inc;
for(std::size_t i = 0; i < vec_size; i += inc)
{
b_type avec = xsimd::load_unaligned(&a[i]);
b_type bvec = xsimd::load_unaligned(&b[i]);
b_type rvec = (avec + bvec) / 2;
xsimd::store_unaligned(&res[i], rvec);
// or rvec.store_unaligned(&res[i]);
}
// Remaining part that cannot be vectorize
for(std::size_t i = vec_size; i < size; ++i)
{
res[i] = (a[i] + b[i]) / 2;
}
}
Aligned vs unaligned memory¶
In the previous example, you may have noticed the load_unaligned/store_unaligned
functions. These
are meant for loading values from contiguous dynamically allocated memory into SIMD registers and
reciprocally. When dealing with memory transfer operations, some instructions sets required the memory
to be aligned by a given amount, others can handle both aligned and unaligned modes. In that latter case,
operating on aligned memory is always faster than operating on unaligned memory.
xsimd provides an aligned memory allocator which follows the standard requirements, so it can be used with STL containers. Let’s change the previous code so it can take advantage of this allocator:
#include <cstddef>
#include <vector>
#include "xsimd/xsimd.hpp"
using vector_type = std::vector<double, XSIMD_DEFAULT_ALLOCATOR(double)>;
void mean(const vector_type& a, const vector_type& b, vector_type& res)
{
using b_type = xsimd::simd_type<double>;
std::size_t inc = b_type::size;
std::size_t size = res.size();
// size for which the vectorization is possible
std::size_t vec_size = size - size % inc;
for(std::size_t i = 0; i < vec_size; i += inc)
{
b_type avec = xsimd::load_aligned(&a[i]);
b_type bvec = xsimd::load_aligned(&b[i]);
b_type rvec = (avec + bvec) / 2;
xsimd::store_unaligned(&res[i], rvec);
// or rvec.store_unaligned(&res[i]);
}
// Remaining part that cannot be vectorize
for(std::size_t i = vec_size; i < size; ++i)
{
res[i] = (a[i] + b[i]) / 2;
}
}
Memory alignment and tag dispatching¶
You may need to write code that can operate on any type of vectors or arrays, not only the STL ones. In that case, you cannot make assumption on the memory alignment of the container. xsimd provides a tag dispatching mechanism that allows you to easily write such a generic code:
#include <cstddef>
#include <vector>
#include "xsimd/xsimd.hpp"
template <class C, class Tag>
void mean(const C& a, const C& b, C& res)
{
using b_type = xsimd::simd_type<double>;
std::size_t inc = b_type::size;
std::size_t size = res.size();
// size for which the vectorization is possible
std::size_t vec_size = size - size % inc;
for(std::size_t i = 0; i < vec_size; i += inc)
{
b_type avec = xsimd::load_simd(&a[i], Tag());
b_type bvec = xsimd::load_simd(&b[i], Tag());
b_type rvec = (avec + bvec) / 2;
xsimd::store_simd(&res[i], rvec, Tag());
}
// Remaining part that cannot be vectorize
for(std::size_t i = vec_size; i < size; ++i)
{
res[i] = (a[i] + b[i]) / 2;
}
}
Here, the Tag
template parameter can be xsimd::aligned_mode
or xsimd::unaligned_mode
. Assuming the existence
of a get_alignment_tag
metafunction in the code, the previous code can be invoked this way:
mean<get_alignment_tag<decltype(a)>>(a, b, res);