common: promote sparse functionality #2328
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
github-actions
bot
added
documentation
spalicki
force-pushed
the
spalicki/sparse_promotion
branch
from
7594c9c to
21ae45b
Compare
spalicki
force-pushed
the
spalicki/sparse_promotion
branch
5 times, most recently
from
9a0e698 to
90e3147
Compare
|
make test |
spalicki
force-pushed
the
spalicki/sparse_promotion
branch
from
90e3147 to
ed97330
Compare
spalicki
changed the title
dzarukin
approved these changes
Mar 3, 2025
spalicki
force-pushed
the
spalicki/sparse_promotion
branch
from
ed97330 to
4f503fa
Compare
|
make test |
densamoilov
reviewed
Mar 4, 2025
densamoilov
reviewed
Mar 4, 2025
densamoilov
reviewed
Mar 4, 2025
densamoilov
reviewed
Mar 4, 2025
| // macro is defined. | ||
| CPU_INSTANCE_SPARSE_X64(jit_uni_sparse_matmul_t) | ||
| CPU_INSTANCE_SPARSE(ref_sparse_matmul_t) | ||
| CPU_INSTANCE_X64(jit_uni_sparse_matmul_t) |
There was a problem hiding this comment.
Now, when the sparse functionality is not experimental, it is possible to pass sparse memory descriptors when creating any primitive and therefore it could lead to undefined behavior because only matmul and reorder primitives have corresponding checks for memory descriptors.
We need to add such checks to other primitives too.
There was a problem hiding this comment.
Just to expand, I think we should check that at primitive_desc create functions for most primitives.
densamoilov
reviewed
Mar 4, 2025
| @@ -75,7 +75,6 @@ dnnl_status_t DNNL_API dnnl_ocl_interop_memory_create(dnnl_memory_t *memory, | |||
| const_dnnl_memory_desc_t memory_desc, dnnl_engine_t engine, | |||
| dnnl_ocl_interop_memory_kind_t memory_kind, void *handle); | |||
|
|
|||
| #ifdef DNNL_EXPERIMENTAL_SPARSE | |||
There was a problem hiding this comment.
We don't have get_mem_object and set_mem_object functions that would take indices. I don't think it's critical, but just need to have that in mind as we may have to add it later on.
densamoilov
reviewed
Mar 4, 2025
vpirogov
reviewed
Mar 4, 2025
vpirogov
reviewed
Mar 4, 2025
vpirogov
reviewed
Mar 4, 2025
spalicki
force-pushed
the
spalicki/sparse_promotion
branch
2 times, most recently
from
8b59600 to
965c9b1
Compare
vpirogov
reviewed
Mar 14, 2025
vpirogov
approved these changes
Mar 14, 2025
spalicki
force-pushed
the
spalicki/sparse_promotion
branch
2 times, most recently
from
363d4bb to
04cdc16
Compare
spalicki
force-pushed
the
spalicki/sparse_promotion
branch
from
04cdc16 to
8904c86
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.
Promoting experimental sparse functionality - common structures and methods. Currently only CPU x64 matmul and reorder have sparse implementations, the rest is guarded by
is_dense_format_kindcall in pd init functions.