Expand description
OpenCL scalar and vector primitive types.
Rust primitives may have subtly different behaviour than OpenCL primitives within kernels. Wrapping is one example of this. Scalar integers within Rust may do overflow checks where in the kernel they do not. Therefore OpenCL-compatible implementations of each of the types are provided so that host and device side operations can be perfectly consistent.
The cl_... (cl_uchar, cl_int, cl_float, etc.) types are simple
aliases of the Rust built-in primitive types and do not behave the
same way that the kernel-side equivalents do. The uppercase-named
types, on the other hand, (Uchar, Int, Float, etc.) are designed
to behave identically to their corresponding types within kernels.
Please file an issue if any of the uppercase-named kernel-mimicking types deviate from what they should (as they are reasonably new this is definitely something to watch out for).
Vector type fields can be accessed using index operations i.e. [0],
[1], [2] … etc. Plans for other ways of accessing fields (such as
.x(), .y(), .s0(), .s15(), etc.) will be considered in the
future (pending a number of additions/stabilizations to the Rust
language). Create an issue if you have an opinion on the matter.
[NOTE]: This module may be renamed.