Convert binary index Hamming callers to dynamic dispatch (#5071)#5071
Open
algoriddle wants to merge 2 commits intofacebookresearch:mainfrom
Open
Convert binary index Hamming callers to dynamic dispatch (#5071)#5071algoriddle wants to merge 2 commits intofacebookresearch:mainfrom
algoriddle wants to merge 2 commits intofacebookresearch:mainfrom
Conversation
Contributor
|
@algoriddle has exported this pull request. If you are a Meta employee, you can view the originating Diff in D100020358. |
…5070) Summary: Convert all 8 public Hamming distance functions in `hamming.cpp` to dynamic dispatch so they use the correct SIMD implementation at runtime instead of always running at scalar speed in DD mode. This is Phase 1 of the Hamming DD conversion. It converts the public API functions in `hamming.cpp`; Phase 2 (future diff) will migrate external callers that use `dispatch_HammingComputer` directly. **Why Hamming is harder than previous DD conversions:** The SIMD code lives inside 11 HammingComputer/GenHammingComputer structs with ISA-specific memory layouts (e.g., NEON `HammingComputer16` stores `uint8x16_t a0` vs x86 `uint64_t a0, a1`). The structs use raw intrinsics (not simdlib) and are template parameters in hot inner loops where `.hamming()` must be inlined. **Approach:** - Extract all template code from `hamming.cpp` into a shared header `hamming_impl.h`, compiled once per ISA TU (AVX2, NEON, NONE). - The `hamdis-inl.h` ISA ladder (`#ifdef __AVX2__` etc.) selects the correct struct definitions based on per-TU compiler flags. - All internal templates live in an anonymous namespace for ODR safety (different TUs see different struct layouts but never share instances). - 8 entry point template specializations (`_dispatch<SL>`) provide the external linkage symbols that the dispatch wrappers call. - `hamming.cpp` includes `hamming_impl.h` with `THE_SIMD_LEVEL=NONE` for the scalar fallback, then dispatches via `with_simd_level_256bit`. - `hamming.h` keeps including `hamdis-inl.h` for backward compatibility with external callers (they continue to work with generic structs). **Functions dispatched:** - `hammings_knn_hc` (binary flat/IVF kNN search) - `hammings_knn_mc` (max-count kNN) - `hamming_range_search` - `hammings` (all-pairs distance) - `generalized_hammings_knn_hc` (byte-level Hamming) - `hamming_count_thres`, `crosshamming_count_thres`, `match_hamming_thres` **Performance impact:** - NEON: HammingComputer16/20/32/64 gain real vectorization (`vcntq_u8`) - AVX2: GenHammingComputer16/32/M8 gain SSE2/AVX2 byte comparison - AVX2: All HammingComputers get hardware `popcnt` (implied by `-mpopcnt`) Differential Revision: D99994593
…earch#5071) Summary: Phase 2 of the Hamming DD conversion. Converts 5 external callers of dispatch_HammingComputer to use per-ISA TUs so the correct HammingComputer structs (AVX2/NEON/generic) are used at runtime. Previously, these files compiled as common TUs with no ISA flags, causing hamdis-inl.h to fall through to generic-inl.h — scalar HammingComputer structs even on AVX2-capable machines in DD mode. Converted files: - IndexBinaryHNSW.cpp: get_distance_computer() factory - IndexBinaryIVF.cpp: get_InvertedListScanner() + search_preassigned() - IndexBinaryHash.cpp: 4 search entry points (knn+range × 2 index types) - IndexIVFSpectralHash.cpp: get_InvertedListScanner() factory - IndexPQ.cpp: polysemous_inner_loop dispatch NOT converted: IndexIVFPQ.cpp polysemous path — deeply interleaved with PQ distance computation inside IVFPQScannerT, niche (off by default, polysemous_ht=0). See ~/ivfpq_polysemous_dd.md for analysis. Differential Revision: D100020358
algoriddle
force-pushed
the
export-D100020358
branch
from
cdd97db to
79d54ee
Compare
meta-codesync
bot
changed the title
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
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:
Phase 2 of the Hamming DD conversion. Converts 5 external callers of
dispatch_HammingComputer to use per-ISA TUs so the correct
HammingComputer structs (AVX2/NEON/generic) are used at runtime.
Previously, these files compiled as common TUs with no ISA flags,
causing hamdis-inl.h to fall through to generic-inl.h — scalar
HammingComputer structs even on AVX2-capable machines in DD mode.
Converted files:
NOT converted: IndexIVFPQ.cpp polysemous path — deeply interleaved
with PQ distance computation inside IVFPQScannerT, niche (off by
default, polysemous_ht=0). See ~/ivfpq_polysemous_dd.md for analysis.
Differential Revision: D100020358