Add floating-point pack/unpack proposal #6191
Conversation
fairywreath
requested review from
tangent-vector,
csyonghe and
expipiplus1
as code owners
| ### Built-in packed Datatypes | ||
| Unlike HLSL's implementation with introduces new packed datatypes, `uint8_t4_packed` and `int8_t4_packed`, unsigned 32-bit integers are used directly | ||
| and no new pakced datatypes are introduced. |
There was a problem hiding this comment.
I don't have a strong preference here and would really appreciate any suggestions or input. If new datatypes are to be introduced we would add something like half2_unorm_packed, half2_snorm_packed, half2_packed, then another full set for the 8-bit variants.
fairywreath
changed the title
| half2 unpack_snorm_f16f16(uint packedVal); | ||
|
|
||
| float2 unpack_half_f16f32(uint packedVal); | ||
| half4 unpack_half_f16f16(uint packedVal); |
There was a problem hiding this comment.
should be half2 here. this should be exactly the same as bitcast<half2>, right?
The proposal currently names them based on HLSL's intrinsics - I believe using Slang's naming convention is better. I will change them and will also add new intrinsics for the integer pack/unpack functions that follows Slang's naming convention in addition to the hlsl intrinsics. |
Related to #6175. Adds a proposal to define floating-point pack/unpack intrinsics.