-
Notifications
You must be signed in to change notification settings - Fork 8
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.
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
Redundant operations in code generated by the TTGT pass #29
Comments
pthomadakis
changed the title
Generated code for TTGT optimization
Redundant operation in code generated by the TTGT pass
Sep 21, 2023
pthomadakis
changed the title
Redundant operation in code generated by the TTGT pass
Redundant operations in code generated by the TTGT pass
Sep 21, 2023
pthomadakis
added a commit
that referenced
this issue
Sep 27, 2023
When one or more tensors are transposed before a contraction (C=A*B), the shape of the contraction result might be different than the expected output C. Thus this intermediate results is stored in a temporary tensor, which will later be tranposed in the C, taking the expected shape. However, instead of just allocating this intermediate tensor we would also transpose tensor C into it, making a redundant trasposition. This commit removes this extra transposition.
pthomadakis
added a commit
that referenced
this issue
Oct 12, 2023
This commit partially addresses issue #29. When one or more tensors are transposed before a contraction (C=A*B), the shape of the contraction result might be different than the expected output C. Thus this intermediate results is stored in a temporary tensor, which will later be tranposed in the C, taking the expected shape. However, instead of just allocating this intermediate tensor we would also transpose tensor C into it, making a redundant trasposition. This commit removes this extra transposition and maintains it only for the case of C+=,-= A*B.
pthomadakis
added a commit
that referenced
this issue
Oct 22, 2023
This commit partially addresses issue #29. When one or more tensors are transposed before a contraction (C=A*B), the shape of the contraction result might be different than the expected output C. Thus this intermediate results is stored in a temporary tensor, which will later be tranposed in the C, taking the expected shape. However, instead of just allocating this intermediate tensor we would also transpose tensor C into it, making a redundant trasposition. This commit removes this extra transposition and maintains it only for the case of C+=,-= A*B.
pthomadakis
added a commit
that referenced
this issue
Oct 22, 2023
This commit partially addresses issue #29. When one or more tensors are transposed before a contraction (C=A*B), the shape of the contraction result might be different than the expected output C. Thus this intermediate results is stored in a temporary tensor, which will later be tranposed in the C, taking the expected shape. However, instead of just allocating this intermediate tensor we would also transpose tensor C into it, making a redundant trasposition. This commit removes this extra transposition and maintains it only for the case of C+=,-= A*B.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
The code generated by the TTGT pass can, in some cases, create redundant transpose operations.
The text was updated successfully, but these errors were encountered: