Bug 177874 – SPIR-V doesn't have any 8-bit types, so WSL can't either

• Component: New Bugs
• Status: RESOLVED WONTFIX

[Myles C. Maxfield]

SPIR-V doesn't have any 8-bit types, so WSL can't either

[Myles C. Maxfield]

Created attachment 322680 [details]
Patch

[Myles C. Maxfield]

According to https://msdn.microsoft.com/en-us/library/bb509646(v=vs.85).aspx and https://www.khronos.org/registry/spir-v/specs/1.2/SPIRV.html#_a_id_capability_a_capability

- HLSL doesn't have 64-bit int types, and they're optional in SPIR-V
- HLSL requires half-precision floating point numbers, and they're optional in SPIR-V
- HLSL requires double-precision floating point numbers, and they're optional in SPIR-V (and already implemented in WSL)

[Filip Pizlo]

(In reply to Myles C. Maxfield from comment #0)
> SPIR-V doesn't have any 8-bit types, so WSL can't either

I think you should be a bit more precise about this.

If it was merely that SPIR-V didn't support 8-bit types, then we should just say that we lower 8-bit types to 32-bit types on the fly.  This is trivial:

- Any 8-bit math operation is expressible in terms of 32-bit math operations.  Most 8-bit math operations lower to exactly one 32-bit math operation and then you ignore the high bits.  The only exceptions are right shifts, division, and modulo.

- 8-bit fields in memory can be packed into 32-bit fields, and then extracted using shifts and masking.

So, it's not correct to say that WSL "can't" have 8-bit types.  It totally can.

Maybe we don't want it to have 8-bit types, but then that's a different issue.

[Myles C. Maxfield]

Migrated to https://github.com/gpuweb/WHLSL/issues/165

[Maciej Stachowiak]

Comment on attachment 322680 [details]
Patch

Unflagging and obsoleting, since this has been WONTFIXEd.