Add int8_t / uint8_t to the language

[AdamJMiles]

Hi,

While this may/may not come naturally with other endeavours in the land of HLSL, I'd like to keep this on the radar.

HLSL currently lacks an 8-bit datatype both for arithmetic and for storage. The primary motivation for having an 8-bit datatype is not for general-purpose arithmetic, but rather:

1. Support storage of smaller types in structures (e.g. StructuredBuffer) and in groupshared memory where space is always at a premium.
2. Ease the sharing of code between C/C++ and HLSL.

It's very difficult to safely load, store and manipulate 8-bit datatypes in groupshared memory without resorting to bitwise operations, masking and even atomics to affect only a single byte of a uint16_t or uint type. CUDA supports char and unsigned char datatypes in its language.

Whether the 8-bit datatypes have any operators (such as +, -, /, * etc) depends on what the level of support for that is in the ecosystem, but Load/Store is the primary motivator here.

Thanks,

Adam

[samuelalonso96]

i would also mention to support 8-bit arithmetic for certain kinds of NNs

[didito]

see #538