Dandandan opened a new pull request, #22096: URL: https://github.com/apache/datafusion/pull/22096
## Which issue does this PR close? - N/A. ## Rationale for this change Primitive group-by vectorized equality had the same non-null comparison loop emitted for both nullable and non-nullable const-generic instantiations. Sharing that loop reduces generated code size and instruction-cache pressure in aggregation paths without changing behavior. Measured with release-nonlto on this branch: - libdatafusion_physical_plan.rlib: 18,068,448 -> 17,592,688 bytes, -475,760 bytes / -2.63%. - rlib text from /usr/bin/size: 4,844,256 -> 4,612,079, -232,177 / -4.79%. - cargo asm aggregate sample: 492,394 -> 464,167, -28,227 / -5.73%. - datafusion-cli __text: 88,257,172 -> 88,249,760, -7,412 bytes. Criterion vectorized_equal_to results across 48 cases showed no statistically significant regressions. Overall mean was -4.94% and median was -1.53%, with larger nullable cases seeing the clearest wins. ## What changes are included in this PR? - Moves the primitive non-null vectorized equality helper out of the NULLABLE const-generic impl and marks it inline(never), so nullable and non-nullable builders share the same emitted helper per primitive type. - Keeps the nullable slow path unchanged. - Updates the aggregate_vectorized benchmark harness to use BooleanBufferBuilder, matching the current GroupColumn API. ## Are these changes tested? - cargo fmt --all - cargo clippy --all-targets --all-features -- -D warnings - cargo test -p datafusion-physical-plan multi_group_by::primitive --profile release-nonlto - cargo bench -p datafusion-physical-plan --features test_utils --bench aggregate_vectorized --profile release-nonlto -- PrimitiveGroupValueBuilder_vectorized_append.*vectorized_equal_to --sample-size 20 --measurement-time 2 ## Are there any user-facing changes? No. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
