Skip to content

Navigation Menu

Sign in
Appearance settings

Search code, repositories, users, issues, pull requests...

Provide feedback

We read every piece of feedback, and take your input very seriously.

Saved searches

Use saved searches to filter your results more quickly

Appearance settings

perf of f16 matmul for small M #2635

Copy link
Copy link
Open
@WilliamTambellini

Description

@WilliamTambellini
Issue body actions

Hello
The perf of f16 2d matmul for small input seems pretty bad:

ArrayFire v3.7.0 (CUDA, 64-bit Linux, build c589451)
Platform: CUDA Toolkit 10.0, Driver: 418.40.04
[0] TITAN V, 12037 MB, CUDA Compute 7.0

M  	K=N	f32	f16
1  	512	0.01383 0.0341257
2  	512	0.01384 0.0341409
4  	512	0.01386 0.0341539
8  	512	0.01385 0.0107975
16 	512	0.01416 0.0108004
32 	512	0.01394 0.0109949
64 	512	0.01510 0.0115619
128	512	0.01784 0.0110581

I guess this is related to M not being multiple of 8.
Anything that could be done to at least reach the perf of f32 ?

Metadata

Metadata

Assignees

No one assigned

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions

      Morty Proxy This is a proxified and sanitized view of the page, visit original site.