Skip to content

Conversation

@V-FEXrt
Copy link
Collaborator

@V-FEXrt V-FEXrt commented Dec 15, 2025

Fixes #7915

There is a detailed writeup in the issue but in summary, native vectors in SM6.9 shouldn't be optimized in the way this path enables and weren't cleanly being rejected. This change now rejects them

Copy link
Contributor

@tex3d tex3d left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I mentioned in the issue here:
I don't understand why vectors should come into play at all, unless there's another bug. Why would you expect to add up the mantissa bits from multiple scalars in a vector to see if it's representable by a scalar int type? Am I missing something?

In the issue:

MinBW for a float2 is 65 so it fails the check and early exits as expected.

But what if it were half? Why is the vector size relevant at all for what should be a scalar transformation?

@V-FEXrt
Copy link
Collaborator Author

V-FEXrt commented Dec 16, 2025

But what if it were half?

For half2 both with and without -enable-16bit-types MinBW is 66 (turns out float2 is also 66 and I had a typo initially)

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

Status: New

Development

Successfully merging this pull request may close these issues.

Invalid floating type error when compiling shader

4 participants