Module columnar 

Columnar execution for high-performance aggregation queries

This module implements column-oriented query execution that avoids materializing full Row objects during table scans, providing 8-10x speedup for aggregation-heavy workloads.

Architecture

Instead of:

TableScan → Row{Vec<SqlValue>} → Filter(Row) → Aggregate(Row) → Vec<Row>

We use:

TableScan → ColumnRefs → Filter(native types) → Aggregate → Row

Benefits

• Zero-copy: Work with &SqlValue references instead of cloning
• Cache-friendly: Access contiguous column data instead of scattered row data
• Type-specialized: Skip SqlValue enum matching overhead for filters/aggregates
• Minimal allocations: Only allocate result rows, not intermediate data

Usage

This path is automatically selected for simple aggregate queries that:

• Have a single table scan (no JOINs)
• Use simple WHERE predicates
• Compute aggregates (SUM, COUNT, AVG, MIN, MAX)
• Don’t use window functions or complex subqueries

Re-exports

pub use batch::ColumnArray;

pub use batch::ColumnarBatch;

pub use filter::apply_columnar_filter;

pub use filter::apply_columnar_filter_simd_streaming;

pub use filter::create_filter_bitmap;

pub use filter::create_filter_bitmap_tree;

pub use filter::evaluate_predicate_tree;

pub use filter::extract_column_predicates;

pub use filter::extract_predicate_tree;

pub use filter::ColumnPredicate;

pub use filter::PredicateTree;

pub use simd_filter::simd_create_filter_mask;

pub use simd_filter::simd_create_filter_mask_auto;

pub use simd_filter::simd_create_filter_mask_packed;

pub use simd_filter::simd_filter_batch;

pub use simd_filter::simd_filter_to_indices;

pub use simd_filter::simd_create_filter_mask_parallel;

pub use simd_filter::simd_filter_batch_parallel;

pub use simd_ops::PackedMask;

Modules

batch

Columnar batch structure for high-performance query execution

filter

Columnar filtering - efficient predicate evaluation on column data

simd_filter

Auto-vectorized filtering for columnar batches

simd_ops

Auto-vectorized SIMD operations for columnar data processing.

Structs

AggregateSpec

A complete aggregate specification

ColumnarScan

Columnar scan over a slice of rows

Enums

AggregateOp

Aggregate operation type

AggregateSource

Source of data for an aggregate - either a simple column or an expression

Functions

can_use_simd_for_column

Determine if a column can use SIMD path based on its data type

columnar_group_by

Compute aggregates with GROUP BY using columnar execution

columnar_group_by_batch

Compute aggregates with GROUP BY using SIMD-accelerated columnar execution

columnar_hash_join_inner

Columnar hash join with SIMD-accelerated key comparison (INNER JOIN)

compute_aggregates_from_batch

Compute aggregates directly from a ColumnarBatch (no row conversion)

compute_multiple_aggregates

Compute multiple aggregates in a single pass over the data

evaluate_expression_to_column

Evaluate an expression on a ColumnarBatch and return the result as a ColumnArray

evaluate_expression_with_cached_column

Evaluate an expression using a cached column as a base value

execute_columnar

Execute a query using columnar processing (AST-based interface)

execute_columnar_aggregate

Execute a columnar aggregate query with filtering

execute_columnar_batch

Execute a columnar query end-to-end on a ColumnarBatch

extract_aggregates

Extract aggregate operations from AST expressions

fast_aggregate_on_rows

Fast single-pass aggregate on rows - avoids batch conversion overhead

simd_aggregate_f64

Compute SIMD aggregate for Float64 columns using streaming batches

simd_aggregate_i64

Compute aggregate for Int64 columns using streaming batches