This library provides functions to do comparions on many rows using AVX2's VPCMP instruction. Fallback code is transparently used if AVX(2) is not available.
The easiest way to explain is to just show the fallback code.
func goVectorEquals[T uint8 | uint16 | uint32 | uint64 | float32 | float64](dst []byte, search T, rows []T) {
clear(dst)
for i, v := range rows {
if search == v {
dst[i/8] |= 1 << (i % 8)
}
}
}Similar to VectorEquals we also implement VectorNotEquals, VectorGreaterThan, VectorLessThan, VectorGreaterEquals and VectorLesserEquals.
goos: linux
goarch: amd64
pkg: github.com/hexon/vectorcmp
cpu: 14th Gen Intel(R) Core(TM) i9-14900
│ purego │ avx2 │
│ sec/op │ sec/op vs base │
VectorEquals8-32 145594.5n ± 2% 628.0n ± 2% -99.57% (p=0.000 n=10)
VectorNotEquals8-32 141155.0n ± 1% 679.7n ± 4% -99.52% (p=0.000 n=10)
VectorGreaterThan8-32 146456.5n ± 2% 856.9n ± 2% -99.41% (p=0.000 n=10)
VectorLessThan8-32 144820.5n ± 1% 858.8n ± 2% -99.41% (p=0.000 n=10)
VectorGreaterEquals8-32 139554.0n ± 2% 928.1n ± 1% -99.33% (p=0.000 n=10)
VectorLesserEquals8-32 141419.5n ± 1% 917.0n ± 3% -99.35% (p=0.000 n=10)
VectorEquals16-32 151.078µ ± 1% 1.325µ ± 2% -99.12% (p=0.000 n=10)
VectorNotEquals16-32 139.751µ ± 3% 1.359µ ± 4% -99.03% (p=0.000 n=10)
VectorGreaterThan16-32 150.466µ ± 2% 1.682µ ± 1% -98.88% (p=0.000 n=10)
VectorLessThan16-32 151.316µ ± 1% 1.700µ ± 2% -98.88% (p=0.000 n=10)
VectorGreaterEquals16-32 139.931µ ± 2% 1.728µ ± 2% -98.77% (p=0.000 n=10)
VectorLesserEquals16-32 138.835µ ± 1% 1.760µ ± 3% -98.73% (p=0.000 n=10)
VectorEquals32-32 146.441µ ± 2% 2.626µ ± 1% -98.21% (p=0.000 n=10)
VectorEqualsFloat32-32 153.134µ ± 1% 2.215µ ± 1% -98.55% (p=0.000 n=10)
VectorNotEquals32-32 140.962µ ± 2% 2.663µ ± 2% -98.11% (p=0.000 n=10)
VectorNotEqualsFloat32-32 160.325µ ± 2% 2.195µ ± 1% -98.63% (p=0.000 n=10)
VectorGreaterThan32-32 144.302µ ± 3% 3.333µ ± 2% -97.69% (p=0.000 n=10)
VectorGreaterThanFloat32-32 146.653µ ± 3% 2.209µ ± 2% -98.49% (p=0.000 n=10)
VectorLessThan32-32 143.874µ ± 2% 3.348µ ± 3% -97.67% (p=0.000 n=10)
VectorLessThanFloat32-32 147.214µ ± 1% 2.221µ ± 1% -98.49% (p=0.000 n=10)
VectorGreaterEquals32-32 139.414µ ± 2% 3.506µ ± 0% -97.49% (p=0.000 n=10)
VectorGreaterEqualsFloat32-32 151.919µ ± 1% 2.226µ ± 1% -98.54% (p=0.000 n=10)
VectorLesserEquals32-32 142.315µ ± 3% 3.447µ ± 3% -97.58% (p=0.000 n=10)
VectorLesserEqualsFloat32-32 152.132µ ± 2% 2.188µ ± 1% -98.56% (p=0.000 n=10)
VectorEquals64-32 146.577µ ± 1% 4.468µ ± 3% -96.95% (p=0.000 n=10)
VectorEqualsFloat64-32 152.734µ ± 2% 4.426µ ± 2% -97.10% (p=0.000 n=10)
VectorNotEquals64-32 141.699µ ± 2% 4.506µ ± 5% -96.82% (p=0.000 n=10)
VectorNotEqualsFloat64-32 164.610µ ± 2% 4.443µ ± 1% -97.30% (p=0.000 n=10)
VectorGreaterThan64-32 146.472µ ± 2% 6.420µ ± 2% -95.62% (p=0.000 n=10)
VectorGreaterThanFloat64-32 149.198µ ± 2% 4.423µ ± 1% -97.04% (p=0.000 n=10)
VectorLessThan64-32 145.543µ ± 2% 6.190µ ± 3% -95.75% (p=0.000 n=10)
VectorLessThanFloat64-32 146.221µ ± 2% 4.373µ ± 1% -97.01% (p=0.000 n=10)
VectorGreaterEquals64-32 139.420µ ± 2% 6.651µ ± 2% -95.23% (p=0.000 n=10)
VectorGreaterEqualsFloat64-32 151.540µ ± 2% 4.426µ ± 1% -97.08% (p=0.000 n=10)
VectorLesserEquals64-32 139.964µ ± 1% 6.508µ ± 3% -95.35% (p=0.000 n=10)
VectorLesserEqualsFloat64-32 151.295µ ± 1% 4.482µ ± 2% -97.04% (p=0.000 n=10)
VectorIsNaNFloat32-32 23.555µ ± 1% 2.228µ ± 1% -90.54% (p=0.000 n=10)
VectorIsNaNFloat64-32 23.687µ ± 1% 4.410µ ± 1% -81.38% (p=0.000 n=10)
geomean 133.0µ 2.492µ -98.13%