Adding ttnn.to_dtype op in TTNN dialect #2055
Open
Conversation
This file contains 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
sdjordjevicTT
requested review from
svuckovicTT,
mtopalovicTT,
jnie-TT,
kmabeeTT,
AleksKnezevic,
pilkicTT,
jserbedzijaTT,
azecevicTT,
tapspatel and
nsmithtt
as code owners
sdjordjevicTT
commented
Jan 31, 2025
| } | ||
|
|
||
| // If the input tensor is on device, we can cast it on the device. | ||
| return this->createDataTypeCastingOp<ttnn::TypecastOp>(op, rewriter, |
There was a problem hiding this comment.
Add assert here that inputLayout.getLayout() must be Tile in this case.
sdjordjevicTT
force-pushed
the
sdjordjevic/add_to_dtype_op_in_ttnn
branch
from
6a219f2 to
1f9d5af
Compare
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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Outdated suggestions cannot be applied.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.
In this pull request (PR), I am introducing the
ttnn.to_dtypeoperation for data type casting on the host within the TTNN dialect.The existing
ttnn.typecastfunction only supports data type casting on a device, which has led to failures when casting on the host. To address this, I have also restructured theTTNNDecomposeLayouts.cppfile to accommodate these changes. Additionally, I have included compiler and silicon tests for scenarios where the inputs are located on the host.Since we are now managing this behavior within the compiler, there is no longer a need for runtime workarounds. As a result, I have removed those workarounds in this PR.
This improvement will benefit
emitC, which previously lacked a runtime workaround and caused several tests to fail. Now, with these changes, we can ensure better compatibility and functionality.I created tests for Silicon from the transform pass. Please review it, and if anyone has a better method for running this on Silicon, your input would be greatly appreciated. Comments are welcome! :)
With this change, we are fixing a couple of bugs noticed:
Closes #1938
Closes #1733
Closes #1658