Hand-tuned SIMD assembly kernels for analytics hot paths.
Reference implementations of performance-critical primitives used in market data processing and statistical computation, written in x86_64 assembly with AVX2. These are called from the higher-level analytics library.
| Kernel | Description |
|---|---|
sum_f64_avx2 |
Horizontal sum of a double-precision vector |
dot_f64_avx2 |
Dot product of two double-precision vectors |
scale_f64_avx2 |
Multiply a vector by a scalar in place |
minmax_f64_avx2 |
Min and max of a vector in a single pass |
nasm -f elf64 src/sum_f64_avx2.asm -o build/sum.o
nasm -f elf64 src/dot_f64_avx2.asm -o build/dot.oAssembled with NASM 2.15+. Target CPU requires AVX2 (Haswell or newer).
Internal component. Reference implementations; production deployments use tuned forks.