Module simd_filter 

SIMD-accelerated filter kernels returning u64 bitmasks.

Each kernel compares a column slice against a scalar and returns a packed Vec<u64> where bit i is set iff the predicate holds for element i. One u64 word covers 64 rows.

Runtime CPU detection selects the fastest path:

• AVX-512 (512-bit, 16 u32 / 8 f64|i64 per op)
• AVX2 (256-bit, 8 u32 / 4 f64|i64 per op)
• Scalar fallback (auto-vectorized by LLVM)

Companion helpers: popcount, bitmask_and, bitmask_or, bitmask_to_indices.

Structs

FilterSimdRuntime

SIMD runtime for filter-to-bitmask operations.

Functions

bitmask_all

Create an all-ones bitmask for row_count rows.

bitmask_and

Bitwise AND of two equal-length bitmasks, SIMD-accelerated.

bitmask_not

Bitwise NOT of a bitmask (flips all bits up to row_count).

bitmask_or

Bitwise OR of two bitmasks, SIMD-accelerated. If the slices differ in length, the longer tail is copied as-is.

bitmask_to_indices

Expand bitmask to a selection vector of row indices.

filter_runtime

Get the global filter SIMD runtime.

popcount

Count total set bits across a bitmask.

words_for

Number of u64 words needed for row_count bits.