[LAYOUTS] Implement toLinearLayout for TensorMemoryEncodingAttr #7748
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
lezcano
changed the title
Jokeren
reviewed
Aug 2, 2025
This `toLinearLayout` is a bit different to way we construct the layouts for SharedEncoding. In particular, here we map the tensor into the memory, and not the other way around. This is to be able to model the `packed=True` version of the layout, where we map two different elements to the same `M/N` location. Representing it this way is not an issue in practice, as we always use these layouts by composing their inverse with a distributed layout, so this way we simply have the inverse already at hand.
|
Different approach after discussing it with @ThomasRaoux. Now we implement it as a map from hardware to the tensor as all the other layouts, which makes everything simpler. Updated the OP reflecting this |
Jokeren
approved these changes
Aug 4, 2025
ThomasRaoux
reviewed
Aug 4, 2025
Comment on lines
+1193
to
+1195
| // We model packed layouts as having the rows/cols dimensions of bitwidth=16 | ||
| // This means that a layout with unpacked=True is the same as one with | ||
| // unpacked=False |
There was a problem hiding this comment.
I think if we will want to track at the byte granularity. For scales we do have 8bits data in the Tensor memory so I think that will help want handling this
There was a problem hiding this comment.
Let's revisit this once we do the scales, but sounds like a reasonable ask
peterbell10
added a commit
that referenced
this pull request
Aug 6, 2025
ptillet
pushed a commit
that referenced
this pull request
Aug 7, 2025
We do so by modelling M/N as describing elements and not the hardware 32bit registers. This allows us to avoid the issue of having two elements pointing to the same register when `unpacked=False`. We also tighten the `MemDescType` verifier and the `TensorMemoryEncodingAttr` verifier to be consistent with the definition we are using. Doing this makes us having to update a ton of lit tests that were silently wrong...
ptillet
pushed a commit
that referenced
this pull request
Aug 7, 2025
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
We do so by modelling M/N as describing elements and not the hardware 32bit registers.
This allows us to avoid the issue of having two elements pointing to the same register when
unpacked=False.We also tighten the
MemDescTypeverifier and theTensorMemoryEncodingAttrverifier to be consistent with the definition we are using. Doing this makes us having to update a ton of lit tests that were silently wrong...