Skip to content

This issue was moved to a discussion.

You can continue the conversation there. Go to discussion →

higher precision reduce #1518

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.

Sign up for GitHub

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

Jump to bottom
Closed
jcupitt opened this issue Jan 7, 2020 · 1 comment
Closed

higher precision reduce #1518

jcupitt opened this issue Jan 7, 2020 · 1 comment
Labels
enhancement

Comments

@jcupitt
Copy link
Member

jcupitt commented Jan 7, 2020

The vector path currently uses a 16 x 16 -> 32 mul followed by a shift. We should experiment with 32 x 32 -> top 32 of 64. We could use the same precision as the C path.

See:

libvips/pyvips#148 (comment)

Similarly for the conv vector path.
@homm
Copy link

homm commented Jan 9, 2020

The vector path currently uses a 16 x 16 -> 32 mul followed by a shift

It's not clear for me how this fits in the previous statement «2.6 bits for the coefficients, 10.6 for the accumulator». 16 x 16 -> 32 is totally enough for 8-bit sources, while «2.6 bits for the coefficients» is not.

@kleisauke kleisauke mentioned this issue May 11, 2020
Merged
@kleisauke kleisauke mentioned this issue Jul 17, 2020
Closed
@jcupitt jcupitt closed this as completed Feb 17, 2021
@libvips libvips locked and limited conversation to collaborators Feb 17, 2021

This issue was moved to a discussion.

You can continue the conversation there. Go to discussion →

Assignees
No one assigned
Labels
enhancement
Projects
None yet
Milestone
No milestone
Development

No branches or pull requests
2 participants
@homm @jcupitt