-
Notifications
You must be signed in to change notification settings - Fork 14
fcvt.bf16.s encoding collides with fround.h from zfa #33
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Comments
Just a ping on this. |
Thanks for pointing this out Alex. I'm looking into a new encoding for fcvt.bf16.s. |
fcvt.bf16.s is now encoded like FCVT.H.S, but with rs2=8 |
@kdockser Could I ask you to roll a new release artifact with this change and a new version number? Having a cite-able PDF with a distinct version will make it easier to do change tracking for downstream projects. In particular, this should unblock us landing experimental support in LLVM. |
Reopened until the change is pushed (which will be soon). |
I see the change is pushed now (thanks!), but a new PDF build and an incremented spec version number would be much appreciated. I could commit the experimental support for bfloat16 with reference to the current commit id, but it's not ideal that people wanting to review the implementation can't easily access a PDF. |
The spec number is now incremented (thanks!) - would it be possible to get a new PDF release? |
A new PDF has been released. This includes that changes suggested by the AR Committee. |
@kdockser I'm not seeing the new PDF? The last tagged release is https://github.com/riscv/riscv-bfloat16/releases/tag/20230322 (March 22nd) |
Another ping, as the PDF still doesn't seem to be posted. Many thanks. |
It seems a new PDF was posted last week, so this request can be considered answered. Thank you. |
With the PDF now available, I've been able to update the LLVM assembler/disassembler patches based on the v0.6 version of the spec:
|
The Zfa extension describes
fround.h
as "encoded like FCVT.H.S, but with rs2=4". This collides with the proposed fcvt.bf16.s encoding which also uses 4 in the rs2 position:The text was updated successfully, but these errors were encountered: