Add EmbeddingSpMDMNBitRowWiseSparse autovectorized variant #5244
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
Contributor
MatzeB
force-pushed
the
export-D89086019
branch
from
403af1e to
96f7476
Compare
MatzeB
added a commit
to MatzeB/FBGEMM
that referenced
this pull request
Dec 19, 2025
) Summary: X-link: facebookresearch/FBGEMM#2235 This adds an autovectorized implementation of the `GenerateEmbeddingSpMDMNBitRowWiseSparse` kernels. This first version is around 10x improvement compared to the reference C++ implementation, but slower than the asmjit verison (which is x86 only), see list below for how to close this gap: Room for further enhacements (see some of the surrounding kernels that do implement them): - This is only specialized for `bit_rate` right now. Specializing for common block sizes typically nets good improvements. - Directly operates on the `out` buffer. This is good for unknown block-sizes, but if we specialize for fixed small block-sizes then a separate buffer is better as it can be promoted completely to vector registers (for fixed vector register size anyway, doesn't work for variable size AArch64 SVE registers). - No prefetching logic yet. Differential Revision: D89086019
MatzeB
force-pushed
the
export-D89086019
branch
from
96f7476 to
91df22b
Compare
MatzeB
added a commit
to MatzeB/FBGEMM
that referenced
this pull request
Dec 19, 2025
) Summary: X-link: facebookresearch/FBGEMM#2235 This adds an autovectorized implementation of the `GenerateEmbeddingSpMDMNBitRowWiseSparse` kernels. This first version is around 10x improvement compared to the reference C++ implementation, but slower than the asmjit verison (which is x86 only), see list below for how to close this gap: Room for further enhacements (see some of the surrounding kernels that do implement them): - This is only specialized for `bit_rate` right now. Specializing for common block sizes typically nets good improvements. - Directly operates on the `out` buffer. This is good for unknown block-sizes, but if we specialize for fixed small block-sizes then a separate buffer is better as it can be promoted completely to vector registers (for fixed vector register size anyway, doesn't work for variable size AArch64 SVE registers). - No prefetching logic yet. Reviewed By: excelle08 Differential Revision: D89086019
MatzeB
added a commit
to MatzeB/FBGEMM
that referenced
this pull request
Dec 19, 2025
) Summary: X-link: facebookresearch/FBGEMM#2235 This adds an autovectorized implementation of the `GenerateEmbeddingSpMDMNBitRowWiseSparse` kernels. This first version is around 10x improvement compared to the reference C++ implementation, but slower than the asmjit verison (which is x86 only), see list below for how to close this gap: Room for further enhacements (see some of the surrounding kernels that do implement them): - This is only specialized for `bit_rate` right now. Specializing for common block sizes typically nets good improvements. - Directly operates on the `out` buffer. This is good for unknown block-sizes, but if we specialize for fixed small block-sizes then a separate buffer is better as it can be promoted completely to vector registers (for fixed vector register size anyway, doesn't work for variable size AArch64 SVE registers). - No prefetching logic yet. Reviewed By: excelle08 Differential Revision: D89086019
MatzeB
force-pushed
the
export-D89086019
branch
from
91df22b to
043def2
Compare
MatzeB
added a commit
to MatzeB/FBGEMM
that referenced
this pull request
Dec 19, 2025
) Summary: X-link: facebookresearch/FBGEMM#2235 This adds an autovectorized implementation of the `GenerateEmbeddingSpMDMNBitRowWiseSparse` kernels. This first version is around 10x improvement compared to the reference C++ implementation, but slower than the asmjit verison (which is x86 only), see list below for how to close this gap: Room for further enhacements (see some of the surrounding kernels that do implement them): - This is only specialized for `bit_rate` right now. Specializing for common block sizes typically nets good improvements. - Directly operates on the `out` buffer. This is good for unknown block-sizes, but if we specialize for fixed small block-sizes then a separate buffer is better as it can be promoted completely to vector registers (for fixed vector register size anyway, doesn't work for variable size AArch64 SVE registers). - No prefetching logic yet. Reviewed By: excelle08 Differential Revision: D89086019
MatzeB
force-pushed
the
export-D89086019
branch
from
043def2 to
af92425
Compare
) Summary: X-link: facebookresearch/FBGEMM#2235 This adds an autovectorized implementation of the `GenerateEmbeddingSpMDMNBitRowWiseSparse` kernels. This first version is around 10x improvement compared to the reference C++ implementation, but slower than the asmjit verison (which is x86 only), see list below for how to close this gap: Room for further enhacements (see some of the surrounding kernels that do implement them): - This is only specialized for `bit_rate` right now. Specializing for common block sizes typically nets good improvements. - Directly operates on the `out` buffer. This is good for unknown block-sizes, but if we specialize for fixed small block-sizes then a separate buffer is better as it can be promoted completely to vector registers (for fixed vector register size anyway, doesn't work for variable size AArch64 SVE registers). - No prefetching logic yet. Reviewed By: excelle08 Differential Revision: D89086019
MatzeB
force-pushed
the
export-D89086019
branch
from
af92425 to
d00772a
Compare
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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.
Summary:
X-link: https://github.com/facebookresearch/FBGEMM/pull/2235
This adds an autovectorized implementation of the
GenerateEmbeddingSpMDMNBitRowWiseSparsekernels.This first version is around 10x improvement compared to the reference C++ implementation, but slower than the asmjit verison (which is x86 only), see list below for how to close this gap:
Room for further enhacements (see some of the surrounding kernels that do implement them):
bit_rateright now. Specializing for common block sizes typically nets good improvements.outbuffer. This is good for unknown block-sizes, but if we specialize for fixed small block-sizes then a separate buffer is better as it can be promoted completely to vector registers (for fixed vector register size anyway, doesn't work for variable size AArch64 SVE registers).Differential Revision: D89086019