Initial proposal for hlsl::numeric_limits (#2)

This seeks to remove the HLSL `#INF` syntax in favor of a more C++-like
hlsl::numeric_limits construct.

proposals/0003-numeric-constants.md

@@ -0,0 +1,47 @@
+# Numeric Constants
+
+* Proposal: [0003](0003-numeric-constants.md)
+* Author(s): [Chris Bieneman](https://github.com/llvm-beanz)
+* Sponsor: [Chris Bieneman](https://github.com/llvm-beanz)
+* Status: **Under Consideration**
+* Planned Version: 202x
+
+## Introduction
+
+A new C++ STL-styled numeric constants interface will enable migrating off older
+inconsistent language features and make cleaner more concise code.
+
+## Motivation
+
+The HLSL Infinity constant syntax violates C token rules by using the `#`
+character, and is not consistent with C or C++ language styling. This causes
+technical challenges by using reserved preprocessor tokens, but also makes the
+feature foreign and less approachable in a C-like language.
+
+## Proposed solution
+
+Starting with adoption the `#INF` token becomes unsupported preferring instead a
+new library class modeled after the C++ `std::numeric_limits` classes. Below is
+the proposed interface for the `hlsl::numeric_limits` class:
+
+```c++
+  template<typename Ty>
+  class numeric_limits {
+  public:
+    static Ty min();
+    static Ty max();
+    static Ty lowest();
+    static Ty denorm_min();
+
+    // Implement infinity in terms of __builtin_huge_val
+    static Ty infinity();
+    static Ty negative_infinity();
+
+    static Ty quiet_NaN();
+    static Ty signaling_NaN();
+  }
+```
+
+This enables a straightforward code transformation from `1.#INF` to
+`hlsl::numeric_limits<float>::infinity()`. While the later is more verbose, it
+is more clear and provides a more consistent C++ style.