feat[dace][next]: Adjusted the handling of non standard GPU Memlets. #1913
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
Once [PR#1976](spcl/dace#1976) is merged in DaCe the code generator is able to handle more Memlets directly as Cuda `memcpy()` calls. This PR modifies the GPU transformation of GT4Py in such a way that these Memlets are no longer transformed into Maps. However, it can only be merged if the DaCe dependency was bumped to a version that includes PR#1976!
edopao
reviewed
Mar 18, 2025
| else: | ||
| is_fortran_order = src_strides[0] == 1 and dst_strides[0] == 1 | ||
| is_c_order = src_strides[-1] == 1 and dst_strides[-1] == 1 | ||
| if is_c_order or is_fortran_order: | ||
| continue | ||
| elif dims > 2: | ||
| if not (src_strides[-1] != 1 or dst_strides[-1] != 1): |
There was a problem hiding this comment.
I was wondering why we do not handle the case dims > 2. Is it because the generic cudaMemcpy() does not provide parameters wOffset and hOffset?
There was a problem hiding this comment.
By handle, I mean why we do not consider the Fortran order.
There was a problem hiding this comment.
This has to do with the code generator that can only C order at the moment for more than 3 dimensions.
This is something I want to fix in DaCe but was not pressing enough yet.
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Outdated suggestions cannot be applied.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.
Once PR#1976 is merged in DaCe the code generator is able to handle more Memlets directly as Cuda
memcpy()calls. This PR modifies the GPU transformation of GT4Py in such a way that these Memlets are no longer transformed into Maps.However, it can only be merged if the DaCe dependency was bumped to a version that includes PR#1976!