Intel® oneAPI DPC++/C++ Compiler Developer Guide and Reference

ID 767253
Date 3/31/2023
Public

A newer version of this document is available. Customers should click here to go to the newest version.

Document Table of Contents

Bounds

Description

bounds_t<LowerT, UpperT> holds the lower and upper bounds of a half-open interval. It is templated to allow the different integer representations for the lower and upper bounds. The intent is to model a valid iteration space over a single dimension.

Bounds can be used to iterate over an entire extent or to restrict iteration space within an extent

Creating Bounds

Bounds can be created using full bounds_t type, but this may be tedious.

bounds_t<int, int>(start, finish)
bounds_t<int, aligned<16>>(start, aligned<16>(finish))
bounds_t<fixed<0>, fixed<1920>>()

It is simpler and clearer to use factory function bounds to build a bounds_t<>.

bounds(start,finish);
bounds(start, aligned<16>(finish));
bounds(0_fixed, 1920_fixed)
Discovering Bounds

Accessors know their valid iteraton space. Initial bounds for an accessor are set to set the lower bound to be fixed<0> and the upper bound set to the value and type of the dimension's extent as specified during construction of the n_container(fixed<>,aligned<>, or int).

To query bounds for given dimension of the accessor use template function bounds_d<int DimensionT>(object).

auto b0 = bounds_d<0>(ca);
auto b1 = bounds_d<1>(ca);
for (int y = b0.lower(); y < b0.upper(); ++y)
    for (int x = b1.lower(); x < b1.upper(); ++x) {
          RGBAs pixel = ca[y][x];
		 // …
    }

bounds_t can participate in C++11 range-based for loops.

for (auto y: bounds_d<0>(ca))
    for (auto x: bounds_d<1>(ca)) {
          RGBAs pixel = ca[y][x];
		 // …
    }

for (auto y: ca.bounds_d0())
    for (auto x: ca.bounds_d1()) {
          RGBAs pixel = ca[y][x];
		 // …
    }

N-Dimensional Indexes and Bounds

To model index and bounds values over multiple dimensions, respectively the following template classes are provided: n_index_t<…> and n_bounds_t<…> . These are both variadic templates, accepting any number of arguments.

n_index is a generator to simplify creating instances of n_index_t.

n_index[540][960]

n_bounds is a generator to simplify creating instances of n_bounds_t.

n_bounds[bounds(540,1080)][bounds(960,1920)]

Alternatively, n_bounds_t can be defined in terms of a n_index_t and n_extent_t.

n_bounds(n_index[540][960], n_extent[540][960]);

Accessing Subsections

From a container's accessors, a new accessor can be created over a subsection defined by a n_bounds_t.

auto ca = c.const_access();
auto subsect = ca.section(n_bounds[bounds(540, 1080)][bounds(960,1920)]);

The effect is to restrict the results of bounds_d<int Dimension> on the subsection accessor.

You can create a new accessor translated to a different index space.

auto offsetnewSpace = ca.translated_to(n_index[1000][2000]);
auto zeroSpace = ca.translated_to_zero();

Accesses will have a translation applied that maps the n_index back to the lower bounds of the accessor that created it. This allows a smaller container to be reused in a larger index space that is being walked over by blocks, or to move a subsection index space back to the origin.