Skip to content

Add thread_warp_size for Metal device in default target attributes#8202

Merged
jwfromm merged 1 commit intoapache:mainfrom
Deelvin:amalyshe/metal_warp_size
Jun 7, 2021
Merged

Add thread_warp_size for Metal device in default target attributes#8202
jwfromm merged 1 commit intoapache:mainfrom
Deelvin:amalyshe/metal_warp_size

Conversation

@elvin-n
Copy link
Copy Markdown
Contributor

@elvin-n elvin-n commented Jun 7, 2021

Recently new transpose op strategy was added for cuda which use thread_warp_size from default target attr parameters.
Since Metal fallback to cuda, it needs this parameter to be defined.
The exact value of this param is still open question, 16 should be safe enough but probably it will be changed later after experiments and more clarifications to 32.

@jwfromm
Copy link
Copy Markdown
Contributor

jwfromm commented Jun 7, 2021

LGTM, thanks @elvin-n

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

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants