0019 - Size type

Introduction

This proposal introduces hlsl::size_t as a size type to represent sizes of objects and offsets into memory.

Motivation

HLSL does not define a size type, and instead uses (generally) uint. This precludes the ability of APIs to represent 64-bit offsets without redefining APIs.

Proposed solution

HLSL should introduce a new hlsl::size_t type which is an implementation-defined “size type”, which must be an unsigned integer large enough to represent the size of any valid object or offset.

APIs in the HLSL standard library should be updated to treat size_t as the type for sizes and offsets.

This change should have no behavior change except in DXC where the sizes in some template arguments were defined as int instead of uint (see: #67).

Detailed design

Types [Math.Types]

The HLSL standard library includes a set of type aliases. The declarations for the aliases shall appear both in the hlsl namespace and in the global namespace.

namespace hlsl {
  typedef _16-bit signed integer type_ int16_t;  // optional
  typedef _32-bit signed integer type_ int32_t;
  typedef _64-bit signed integer_ int64_t;

  typedef _16-bit unsigned integer type_ uint16_t;  // optional
  typedef _32-bit unsigned integer type_ uint32_t;
  typedef _32-bit unsigned integer type_ uint;
  typedef _64-bit unsigned integer type_ uint64_t;

  typedef _16-bit floating point type_ float16_t; // optional
  typedef _32-bit floating point type_ float32_t;
  typedef _64-bit floating point type_ float64_t;

  typedef _unsigned integer type_ size_t;
}

The signed integer data types defined as intN_t and unsigned integer data types defined as uintN_t specify the specific bit-width of the object and value representation of the object. The data type uint will be a 32-bit unsigned integer type.

The floating point data types defined as floatN_t specify the specific bit-width of the object and value representation of the object, and shall use the corresponding binary formats defined by \gls{IEEE754} (i.e. binary16, binary32, binary64).

The data type size_t will be an implementation-defined unsigned integer type that is large enough to contain the size in bytes of any object.