Documentation
¶
Overview ¶
Package simd provides vectorized kernels for hot columnar loops, each with a portable pure-Go fallback selected at build time by architecture and, on amd64, at runtime by CPU feature. The assembly is generated by avo (internal/cmd/gensimd) and committed (*_amd64.s); the pure-Go path is the reference the assembly is fuzzed against, so results are bit-identical on every platform.
Index ¶
Constants ¶
const EqualFixed16Width = 16
EqualFixed16Width is the fixed row width EqualFixed16 requires: both blob's rows and needle must be exactly this many bytes. It is a hardcoded property of the kernel (the AVX2 path compares a 16-byte needle broadcast into a 32-byte vector), not a parameter — a caller comparing against len(needle) or some other column's width instead of this constant is asserting the wrong thing, since neither actually guarantees the kernel's real precondition.
Variables ¶
This section is empty.
Functions ¶
func EqualFixed16 ¶ added in v0.32.0
func EqualFixed16(blob, needle, dst []byte)
EqualFixed16 sets dst[i] = 1 where the i-th 16-byte row of blob equals needle, else 0. len(blob) must be len(dst)*16 and len(needle) must be 16. On a CPU with AVX2 and enough rows to amortize dispatch it uses the vector kernel, else the portable reference; both agree bit-for-bit.
func MinMaxFloat64 ¶ added in v0.10.0
MinMaxFloat64 returns the minimum and maximum of s, ignoring NaN. An empty or all-NaN slice returns (+Inf, -Inf) — i.e. min > max — the sentinel a caller reads as "no real values". On a CPU with AVX2 and a slice long enough to amortize the dispatch it uses the vector kernel, else the portable reference; both return identical results.
func MinMaxInt64 ¶
MinMaxInt64 returns the minimum and maximum of s (and 0, 0 for an empty slice). On a CPU with AVX2 and a slice long enough to amortize the dispatch it uses the vector kernel, else the portable reference; both return identical results.
Types ¶
This section is empty.
Source Files