{"id":10888,"date":"2025-07-28T10:10:02","date_gmt":"2025-07-28T10:10:02","guid":{"rendered":"https:\/\/www.modernescpp.com\/?p=10888"},"modified":"2025-07-28T10:10:02","modified_gmt":"2025-07-28T10:10:02","slug":"data-parallel-types-reduction","status":"publish","type":"post","link":"https:\/\/www.modernescpp.com\/index.php\/data-parallel-types-reduction\/","title":{"rendered":"Data-Parallel Types: Reduction"},"content":{"rendered":"\n

In this article, I will discuss reduction and mask reduction for data-parallel types.<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

Reduction<\/h2>\n\n\n\n

A reduction reduces the SIMD vector to a single element. The library provides three functions for this purpose: reduce<\/code>, hmin<\/code>, and hmax<\/code>.
The following program shows how these functions are used.<\/p>\n\n\n\n

\/\/ reduction.cpp<\/span>\n\n#include <array><\/span>\n#include <experimental\/simd><\/span>\n#include <functional><\/span>\n#include <iostream><\/span>\n#include <string_view><\/span>\n\nnamespace<\/span> stdx =<\/span> std::<\/span>experimental;\n \nvoid<\/span> println<\/span>(std::<\/span>string_view name, auto<\/span> const<\/span>&<\/span> a) {\n    std::<\/span>cout <<<\/span> name <<<\/span> ": "<\/span>;\n    for<\/span> (std::<\/span>size_t<\/span> i{}; i !=<\/span> std::<\/span>size(a); ++<\/span>i)\n        std::<\/span>cout <<<\/span> a[i] <<<\/span> ' '<\/span>;\n    std::<\/span>cout <<<\/span> '\\n'<\/span>;\n}\n\n \nint<\/span> main<\/span>() {\n\n    const<\/span> stdx::<\/span>fixed_size_simd<<\/span>int<\/span>, 16<\/span>><\/span> a([](int<\/span> i) { return<\/span> i; });    \n    println("a"<\/span>, a);\n\n    auto<\/span> sum =<\/span> stdx::<\/span>reduce(a);\n    std::<\/span>cout <<<\/span> "sum: "<\/span> <<<\/span> sum <<<\/span> "<\/span>\\n\\n<\/span>"<\/span>;\n\n    const<\/span> stdx::<\/span>fixed_size_simd<<\/span>int<\/span>, 8<\/span>><\/span> b([](int<\/span> i) { return<\/span> i +<\/span> 1<\/span>; });    \n    println("b"<\/span>, b);\n\n    auto<\/span> product =<\/span> stdx::<\/span>reduce(b, std::<\/span>multiplies<><\/span>());\n    std::<\/span>cout <<<\/span> "product: "<\/span> <<<\/span> product <<<\/span>  "<\/span>\\n\\n<\/span>"<\/span>;\n\n    auto<\/span> maximum =<\/span> stdx::<\/span>hmax(b);\n    std::<\/span>cout <<<\/span> "maximum: "<\/span> <<<\/span> maximum <<<\/span>  "<\/span>\\n\\n<\/span>"<\/span>;\n\n    auto<\/span> minimum =<\/span> stdx::<\/span>hmin(b);\n    std::<\/span>cout <<<\/span> "minimum: "<\/span> <<<\/span> minimum <<<\/span>  "<\/span>\\n\\n<\/span>"<\/span>;\n    \n}\n<\/pre><\/div>\n\n\n\n
First, the reduce<\/code> function is used. By default, the + operator is applied in the usual way. However, this function can also be parameterized with any binary operator. The expression stdx::reduce(b, std::multiplies<>()) <\/code>applies the function object std::multiplies <\/code>from the header functional. The functions hmax<\/code> and hmin<\/code> determine the maximum and minimum of the SIMD vector b<\/code>.
Finally, the program output follows:<\/p>\n\n\n\n
\"\"<\/figure>\n\n\n\n
Mask Reduction <\/h2>\n\n\n\n
When reducing a mask, the SIMD mask is reduced to either true <\/code>or false <\/code>value.<\/p>\n\n\n\n
all_of, any_of, none_of, <\/code>and some_of<\/code><\/h3>\n\n\n\n
<\/p>\n\n\n\n
Here we encounter some old friends from C++: all_of, any_of, none_of, <\/code>and some_of<\/code>. When using these functions, the SIMD mask is reduced to either  true <\/code>or false <\/code>value.<\/p>\n\n\n\n