Skip to content
This repository was archived by the owner on Oct 31, 2025. It is now read-only.

Implement some float packing/unpacking instructions#709

Merged
khyperia merged 2 commits intomainfrom
float-packing
Aug 10, 2021
Merged

Implement some float packing/unpacking instructions#709
khyperia merged 2 commits intomainfrom
float-packing

Conversation

@khyperia
Copy link
Contributor

@khyperia khyperia commented Aug 9, 2021

Fixes #707

Wasn't too hot on the GLSL/SPIR-V name of packHalf2x16, it's really confusing to me what the heck that means, so tried to go more explicit with the name - at least to me, f32x2_to_f16x2 is much more clear (and in line with the scalar f32_to_f16 version)

Left off f64 conversion insts for now, ark doesn't need them any time soon and they're weird (I think they're transmute (u32, u32) <-> f64, not an actual bit-changing conversion?)

@khyperia khyperia requested a review from eddyb as a code owner August 9, 2021 09:49
@hrydgard
Copy link
Contributor

hrydgard commented Aug 9, 2021

The new naming is better. Probably descriptive enough while still avoiding SSE-intrinsic-style acronym salad. The u32 types makes it clear enough that it's about packed values.

@khyperia khyperia merged commit 1a3a12f into main Aug 10, 2021
@khyperia khyperia deleted the float-packing branch August 10, 2021 11:54
Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Expose floating point format conversion operations (16 <-> 32 bits)?

2 participants